THE EFFECT OF COLLEGIATE ATHLETICS ON ALUMNI GENEROSITY
Except where reference is made to the work of others, the work described in this thesis is
my own or was done in collaboration with my advisory committee. This thesis does not
include proprietary or classified information.
________________________________________
Christopher Andrew Whaley
Certificate of Approval:
_________________________ _________________________
John D. Jackson Steven B. Caudill, Chair
Professor Regions Bank Professor
Economics Economics
_________________________ _________________________
James E. Long Stephen L. McFarland
Professor Acting Dean
Economics Graduate School
ii
THE EFFECT OF COLLEGIATE ATHLETICS ON ALUMNI GENEROSITY
Christopher Andrew Whaley
A Thesis
Submitted to
the Graduate Faculty of
Auburn University
in Partial Fulfillment of the
Requirements for the
Degree of
Master of Science
Auburn, Alabama
December 15, 2006
iii
THE EFFECT OF COLLEGIATE ATHLETICS ON ALUMNI GENEROSITY
Christopher Andrew Whaley
Permission is granted to Auburn University to make copies of this thesis at its discretion,
upon request of individuals or institutions and at their expense. The author reserves all
publication rights.
______________________________
Signature of Author
______________________________
Date of Graduation
iv
VITA
Christopher Andrew Whaley, son of Ronald Whaley and Virginia Whaley, was
born September 13, 1982, in Cordova, Tennessee. He graduated from Germantown High
School in 2000. Chris attended Auburn University in the fall of 2000 and graduated cum
laude with a Bachelor of Science degree in Business Administration in August, 2004. He
entered Graduate School, Auburn University, that same fall.
v
THESIS ABSTRACT
THE EFFECT OF COLLEGIATE ATHLETICS ON ALUMNI GENEROSITY
Christopher Andrew Whaley
Master of Science, December 15, 2006
(B.S.B.A., AU, 2004)
76 Typed Pages
Directed by Steven Caudill
This study tests the effect collegiate athletics plays within the parameters of a
model for alumni giving. A sample of over 650 schools, both with and without athletic
programs, indicates universities participating in NCAA football can see substantial
increases in alumni revenues, while basketball programs have no significant bearing on
contributions. Further investigation into the relative success of an athletic team proves
inconsequential, concluding that it is not how well a university performs; only that it
performs at all.
vi
ACKNOWLEDGEMENTS
The author would like to thank Dr. Jim Long for the opportunity to pursue
graduate studies in Economics, Dr. John Jackson for having the time to answer questions,
and Dr. Richard Beil for listening to the ideas. Special thanks are also due to Dr. Steven
Caudill for guidance, assistance and endless patience throughout the writing of this paper.
vii
Style manual or journal used: American Journal of Economics
Computer software used: SAS, LIMDEP, Microsoft Excel, Microsoft Word
viii
TABLE OF CONTENTS
LIST OF TABLES .........................................................................................................ix
CHAPTER PAGE
I. INTRODUCTION ............................................................................................1
II. LITERARY REVIEW.......................................................................................4
III. THE DATA ....................................................................................................17
IV. DETERMING THE EFFECT OF ATHLETIC PROGRAMS..........................27
V. DECOMPOSITION........................................................................................35
VI. ESTIMATING THE EFFECT OF ATHLETIC SUCCESS ............................40
VII. CONCLUSION...............................................................................................44
REFERENCES..............................................................................................................47
APPENDIX...................................................................................................................50
ix
LIST OF TABLES
TABLE PAGE
Dependent Variable Information....................................................................................50
Explanatory Variables ...................................................................................................51
Explanatory Variable: Descriptive Statistics ..................................................................52
Mean Values of Dependent Variables ............................................................................53
Linear Regression Results with Differing Dependent Variables .....................................54
Correlation Matrix.........................................................................................................55
Variance Inflation Factors..............................................................................................58
Heteroscedasticity Consistent Estimates ........................................................................59
Regression Results Separated by Football .....................................................................60
Decomposition ..............................................................................................................61
Decomposition II...........................................................................................................62
Descriptive Statistics of Athletic Success Variables.......................................................63
Mean Values of Variables by Division...........................................................................64
Regression Results with Differing Success Variables.....................................................65
F-Statistics on Division Regressions ..............................................................................67
1
CHAPTER I
INTRODUCTION
For over 130 years, academic institutions have participated in intercollegiate
sports. Collegiate athletics have grown into multi-billion dollar industry. In the 2003-
2004 season, total football revenue collected from universities in NCAA Division I
eclipsed $1.5 billion dollars.
1
Clearly, intercollegiate athletics generates sizeable
revenues, but many argue that these programs are indulgences draining valuable
resources from academic programs. Critics say that athletics not only distract students
from academic pursuits, but also divert finances away from the core of the institution.
Empirical results support this claim. For example, as of 2003, spending on Division I
athletic programs rose about 25 percent, while total university spending increased by only
10 percent. In addition, only 40 percent of institutions participating in collegiate athletics
reported self-sufficient athletic programs.
2
Are there benefits from athletic programs? A growing area of research examines
the externalities associated with collegiate athletics. The effects of athletic programs on,
for example, student quality (Mixon, 1995), graduation rates (Tucker, 2004) and
applicant pool sizes (Toma and Cross, 1998) have been recently examined. This thesis
examines the connection between athletic programs and alumni contributions.
1
Office of Post-Secondary Education Equity in Athletics Disclosure Website at
http://ope.ed.gov/athletics/index.asp
2
The findings are based off data released from the NCAA and the U.S. Department of
Education. (Sylwester and Witosky, 2004)
2
In 2005, voluntary support to higher education exceeded $25 billion. Of that
amount, 28 percent (roughly $7 billion) came from alumni.
3
As giving from alumni has
become increasingly important to university finances, many have speculated about the
effect athletic programs have on the generosity of donors. It is easy to see why most
assume that college sports have a positive influence on donation levels. The degree to
which the public is exposed to the Bowl Championship Series and March Madness via
television, media, and the Internet, reinforces a common sense belief that college sports
have the potential to generate enormous revenue for colleges and universities. Others
dismiss this idea, arguing that college-level athletics, particularly basketball and football,
are dominated by a small collection of successful teams who receive the majority of
publicity and revenues. Even so, there are few systematic examinations of the
relationship between athletics and alumni giving.
This thesis presents an empirical analysis of the relationship between collegiate
athletics and alumni generosity. Unlike previous literature, a comprehensive data set of
schools both with and without athletic programs is used. This allows, for the first time, a
complete analysis of athletics and giving, using non-athletic institutions as a benchmark
for comparison. The study further examines alumni contributions not only as a factor of
athletics, but also other important student, institutional and economic characteristics.
This paper includes three sections of specific methodology. First, a contributions
model is subjected to OLS regression analysis to determine the effect of athletic
programs. Simple dummy variables are constructed to account for participation in
NCAA football and basketball. Next, a decomposition technique is used to establish the
3
Information is found at www.cae.org
3
relative differences in schools with and without athletic programs. Finally, the same
contributions model will be used to verify any relationship athletic success measures
might have on giving. The finds show a significant increase in alumni generosity is
associated with schools having athletic programs. In particular, institutions participating
in NCAA football could see on average $3.3 million more in alumni giving. However,
for those institutions with athletics, there is no difference in giving between schools
having ?successful? programs and those having ?unsuccessful? programs.
4
CHAPTER II
LITERATURE REVIEW
There are several theoretical frameworks for the economic modeling of charitable
giving. The first discussed in this thesis views donations as part of consumer theory,
setting contributions as a function of prices and income. Feldstein (1976) developed a
demand model for charitable giving using itemized tax returns gathered from 1948-1968
by the Internal Revenue Service. Itemized charitable contributions were divided into five
major categories, one being educational institutions. The amount of charitable
contributions for education is used as the dependent variable in several regression
models. Feldstein employed the average adjusted gross income per return as the measure
of income and one minus the marginal federal tax rate as the price. Economic theory
suggests that a person contributing one dollar towards charity, because of the deduction,
could have one minus the marginal tax rate of additional consumption. Using a double
log model, the author reported that gifts to educational institutions are extremely sensitive
to price, i.e. marginal tax rates. He finds that eliminating federal deductions on charitable
contributions would reduce those donations by 50 percent.
The second approach to charitable giving involves the individual?s desire to
provide public goods for society. Much of the previous literature focused on explaining
the following: person A gives a donation to person B and both receive a positive level of
utility; however, as B?s income rises (due to the contribution from A), A gives a smaller
5
amount than before.
4
Keating (1981) felt that contributions were not made based on the
relative income of others, but based on a general need to contribute. Keating examined
data collected from 1000 individuals contributing to the United Way in the autumn of
1979. Using a probit model, Keating modeled the probability of giving as a function of
income level and a series of donor characteristics, including the age, education, and sex
of the respective donor. The results indicated the existence of a positive and significant
relationship between age and donations, while education and sex were insignificant.
These two theories explaining generosity provide insight into the development of
a model for philanthropic contributions. The focus here is on a specific instance of
giving, that is, charitable giving to academic institutions. Of particular interest is the
correlation between alumni giving and the presence and success of athletic programs.
Several authors have previously investigated this relationship.
Sigelman and Carter (1979) were among the first to conduct an empirical
investigation in this area. While previous studies focused on how an institution?s
decision to drop football affected alumni giving, Sigelman and Carter recognized that
universities most likely to drop football programs are those with weak football teams that
may not inspire generosity among alumni. The authors also dismissed widely reported
casual empirical evidence. For example, Virginia Tech won the National Invitational
Tournament for men?s basketball in 1973. Afterwards, their president reported increases
in giving in the thousands of dollars and preferential treatment from the legislature.
Sigelman and Carter conclude from an analysis of giving data for years preceding and
succeeding the NIT victory that there was no increase in donations; ?what we have is a
4
See Schwartz (1970)
6
wealth of speculation and a lack of conclusive evidence concerning the input of athletic
success on alumni giving.?(Sigelman and Carter, 287)
Sigelman and Carter also are among the first to attribute alumni generosity not
only to number of alumni but also to social class characteristics of those alumni. Even
so, this idea is not introduced in their empirical work. Using data gathered on 138
Division I ?big time? institutions for the 1975-1976 academic year, Sigelman and Carter
apply several simple correlation and regression techniques. In each analysis, they specify
three measures of alumni generosity: 1) percent change in total alumni giving, 2) percent
change in the dollar value of the average gift, and 3) percent change in the proportion of
alumni who contributed. These three dependent variables were regressed on three
independent variables related to athletic success: football basketball winning percentage,
basketball winning percentage, and a dummy variable equal to one if the institution
participated in a post-season football bowl game. In their simple correlation models,
none of the athletic performance measures proved to have any significant relationship to
alumni giving; in fact, three of the performance measures were negatively correlated with
giving.
5
Sigelman and Carter?s simple regression models gave similar results. None of
the coefficients on the athletic success measures were significant. In fact, the overall
explanatory power of the model (R
2
) was zero. Sigelman and Carter concluded that no
significant relationship between athletic performance and alumni generosity could be
found. The authors did suggest other statistical tests as recommendations for future
papers.
5
Zero of the nine independent variables were significant at a 0.05 level of probability.
7
Sigelman and Carter?s work encouraged many similar efforts. However, their
paper has one major problem not addressed in the literature for some time. Their analysis
omits certain key independent variables that affect alumni generosity such as academic
success. Universities, while excelling on the football field, may also be improving
academically. If measures of academic success are omitted from the regressions, these
effects may be captured by other variables in the regression model, yielding inaccurate
results.
Brooker and Klastorin (1981) recognized some of the limitations of Sigelman and
Carter?s analysis and conducted a reexamination of their data. Brooker and Klastorin
conclude that Sigelman and Carter implicitly assumed that all universities and colleges
are homogeneous. The authors note that Division I schools differ along several
dimensions: public/private, size of student body, alumni numbers, and religious/secular.
Brooker and Klastorin employed a smaller data set than Sigelman and Carter.
Using data from years 1964-1965 through 1971-1972, they selected 58 schools
representing most of the major Division I athletic conferences. The independent
variables selected for use in their regression models are similar to those used by Sigelman
and Carter, but with some additions. Brooker and Klastorin added winning percentage
lagged one and two years, ranking in final UPI Top 20 national poll (football and
basketball), and lagged values of those rankings.
6
GNP was included to incorporate
fluctuations in economic conditions. Three dependent variables were chosen to measure
alumni generosity: size of the average gift, per capita gift to alumni, and percentage of
6
United Press International was the approved ranking system before the modern poll and
BCS system.
8
alumni solicited contributing to annual fund.
7
To further control for institutional
differences, Brooker and Klastorin grouped schools by size and public/private status.
Their findings are summarized in the following quote:
To question whether athletic performance influences alumni giving, contrary to
the generalized conclusion of Sigelman and Carter, the answer seems to be, ?yes,
but it depends on some institutional factors.? (Brooker and Klastorin, 746)
One of the major factors is the nature of the institution; whether public or private. In
every regression for every group of schools, at least one or more of the measures of
athletic success was associated with giving. However, no consistent effect (positive or
negative) was found. For public institutions, only major state universities had significant
relationships between athletic success and giving. There was a negative relationship
between football success and the size of per capita gifts, indicating that the giver is
inclined to give less if the football team is doing well (although the authors do assume
that there is an overall increase in the number of alumni contributing).
Brooker and Klastorin also discuss a hidden benefit of collegiate athletic success,
favorable attitudes from legislatures. This, alone, could lead to increases in funding.
However, they dismiss the investigation citing the difficulty in any accurate
determination. The authors admit the potential of favorable treatment, but warn careful
consideration from individual institutions relying on such an outcome. Their results do
indicate that future analysis should include some distinction between public and private
institutions.
7
Per capita gifts were chosen to account for schools of differing size. Annual funds are
mentioned throughout the paper. These funds are often general giving accounts that go
specifically to the university, including a variety of earmarked and all-purpose gifts.
9
Sigelman and Bookheimer (1983) examined the alumni-athletics relationship,
focusing on contributions specifically earmarked for athletic programs. The authors give
an excellent discussion of the problems associated with collegiate athletic programs.
Sigelman and Bookheimer address issues like poor ticket sales for losing institutions and
ticket price ceilings for schools facing professional sports competition. Their analysis
includes not only an examination of donations specifically to athletic departments, but
also a study of voluntary giving to schools, in general, to determine the relationship
between giving to athletic departments and giving to annual funds.
Sigelman and Bookheimer obtained financial data on various collegiate athletic
programs from a survey conducted by the Omaha World-Herald and published in the
Chronicle of Higher Education (Middleton, 1982). The measurement of athletic success,
usually based on the won/loss record, was calculated and reported as a ?success score.?
These ?success scores? used a linear decay function to weight the four previous year?s
winning percentages, in order to incorporate the lagged effect of athletic performance
into their analysis. This is based on the idea that wins four years ago do not matter as
much as wins today.
Sigelman and Bookheimer are among the first to incorporate socioeconomic
factors into the generosity analysis. The authors include the number of alumni, a
public/private dummy, and an academic quality measure.
8
Also included are three novel
independent variables. The ?Hotbed? variable is an index of sports-craziness. This
variable is measured three ways: percentage of state population engaged in agriculture,
8
The academic quality measure used in Sigelman and Bookheimer is a composite scale
designed by Barron?s Profiles of America Colleges. It is an index ranging from 1 (non-
competitive) to 9 (most competitive) and based on admission selectivity.
10
median educational attainment of state adult population, and median family income of the
state. The ?Prevailing Ethos? recognizes that certain areas promote civic responsibility
and hence a decreased value on non-noble pursuits (for example, donating to an athletic
department). This is measured by the Sharkansky-Elazar political culture index assigning
a score of 1 for the most moralistic areas and 9 for most traditionalistic (likely athletic
contributors). Finally, they use the ?Only Game in Town? factor. This measures the
competition universities face from professional sports teams. It is denoted by a dummy
variable having a ?1? value if the institution is located within a two hour driving distance
from professional football or basketball franchise and a ?0? value if otherwise. In order
to conduct the parallel analysis, total alumni giving is used as a second dependent
variable.
Sigelman and Bookheimer conducted simple correlation analysis between the
aforementioned independent variables and the two dependent variables. They report that
there is no significant relationship between total alumni giving and athletic giving. This
is important to note because it allows the authors to work with two different dimensions
of voluntary support. Only two significant correlations were found. The first was
between traditionalistic nature of a school and athletic contributions. This positive
correlation implies that schools in more traditionalistic areas have less civic desire and
more self-interested wants (i.e. they would rather spend for have a luxury box at a
football game than spend for a scholarship to help an unknown student). The second
correlation is between football success and athletic contributions. Here again, the
positive correlation follows logic; a better team encourages more donations.
11
Sigelman and Bookheimer also estimated several multiple regression models in
order to isolate the effects of performance on contributions. Due to the large number of
independent variables and the small size of the available samples, the authors
encountered problems with degrees of freedom. As a solution, Sigelman and
Bookheimer used a stepwise regression procedure, allowing variables to enter the
regression model only if they significantly increased the model?s explanatory power. A
regression model was developed regressing football success score on athletic
contributions.
9
It was concluded that a one unit increase in the football success score
brought about an increase of $1,251,600, or, more appropriately, for every ten percent
increase in the football success translates into $125,160 more of voluntary support to the
athletic department. Overall, Sigelman and Bookheimer examined a large amount of data
and concluded the following:
In sum, the increasingly precarious financial situation in which most athletic
programs find themselves is apt to beget an emphasis on winning which has many
unfortunate consequences, and also to set into opposition the economic interests
of the largely independent athletic program and those of the rest of campus.
(Sigelman and Bookheimer, 358)
Further, Sigelman and Bookheimer commented on the implications of these finding,
citing that traditional athletic powerhouse wins are loses for doormat schools. Good and
bad football programs still spend money on their teams. Much like certain societies
income gaps, better schools get richer, worse schools get poorer.
Many of the previous studies face problems with model definition and capturing
all the factors that go into alumni generosity. There was never any real discussion of case
studies or institutional investigation. However, in 1994, Grimes and Chressanthis
9
No other variables could meet the criteria once added.
12
published a study done specifically on one university. Through their article deals with
time-series data, relevant information regarding their model is included in this paper.
Their subject was Mississippi State University. Arguing that it is a median
representative of Division I-A schools, Grimes and Chressanthis examined data given by
the Mississippi State Development foundation from 1962-1991.
10
This gave the authors
a two-fold advantage. The previous studies used alumni giving data based off reported
?annual? funds. These funds are reported by law; however, they often include a
hodgepodge of assorted donations that range from general gifts to restricted funds. Also,
there is no confusion between generosity towards the university and donations towards
athletic departments.
The authors were also some of the first to incorporate more detailed sport related
variables in the testing. They included not only television appearances, but also NCAA
sanctions. This was of particularly new interest to see how a group like the NCAA could
affect a school?s earning potential.
Grimes and Chressanthis outlined a need-based approach to alumni giving,
including enrollment and appropriations variables. Larger student populations require
that the university spend more money on the care and instruction of those students.
Coupled with the effect of state appropriations, these variables help determine the need
for alumni donations and hence, the degree to which the institution (i.e. Mississippi State
Development Foundation) will procure those funds. Grimes and Chressanthis also
10
Mississippi State Development Foundation is the main source of alumni contributions
and information concerning Mississippi State University. They distinguish between gifts
of different types of private giving by source and function. MSU possesses a separate
group know as the Bulldog club that receives and disperses money to athletics.
13
included the number of televised games per year (Television). This helps to determine
the relevance of what many authors have called the ?advertising effect of athletics?.
11
MSU participates in several NCAA sports, however, their revenue collecting
programs included football, basketball, and baseball. Grimes and Chressanthis ran three
regressions each using its own respective athletic measures (i.e. one for football, one for
basketball, and one for baseball) and a fourth involving total athletic performance. While
all the regressions reported similar findings on non-athletic variables, the results on
athletics differ in each regression.
12
Under football it was determined that only NCAA
sanctions were significant suggesting that these sanctions cause decreases in alumni
giving. Winning percentage provided the only significant sports variable under the
basketball information. A positive correlation shows that better basketball seasons
induce higher giving. Television appearances proved highly significant in baseball data,
showing a positive relation to alumni donations. Finally, the total regression showed that
both winning percentage and television appears are positively and significantly correlated
with generosity.
13
Grimes and Chressanthis conclude that, overall, a successful athletic program
does possess spillover benefits in the form of increased alumni generosity. However, the
authors do mention some of the limitations of their study. The case study of MSU,
though a median representative, cannot be generalized due to the specific nature of the
11
Advertising effect of athletics, see Bremmer and Kesselring (1993)
12
Appropriations and Income were significant in all four models. Alumni base proved
significant in all three individual sports models, but failed to be substantial in the
collective regression.
13
Coefficients for football sanctions, basketball winning percentage, and overall
television appearances are significant at a 95% level. Estimates on baseball television
appearances are reported at 99% level, while overall winning percentage is at a 90%
level.
14
school. Further, the study only examined the monetary effects of collegiate athletics and
ignored moral views concerning athletics and higher education:
While this analysis cannot possibly attempt to settle all the controversies
surrounding intercollegiate athletics, it has provided some evidence that a major
sports program is more than a series of social events, and can yield external
monetary benefits for academics. (Grimes and Chressanthis, 38)
The message provided here seems to reinforce the principle issued by Brooker and
Klastorin (1981) that the spillover benefits of athletics depend on the institution.
Discussion of the next paper diverges from the athletics and provides a
comprehensive model for alumni donations. Baade and Sundberg (1995) presented a
paper attempting to determine the factors affecting alumni-specific generosity. They
chose to design several models off three sets of variables: institution characteristics,
student characteristics, and efforts to solicit funds. The authors used a dataset of more
than 250 differing institutions over the 1989 and 1990 fiscal years.
14
The data underwent
longitudinal regression analysis producing three separate regressions: public universities,
private universities, and liberal arts colleges. The defining characteristic of their study is
their introduction of ?family wealth?. This idea was to capture the overall wealth of
students entering a university as potential for them to donate. They used a proxy in place
of a direct measurement.
15
Their conclusions on the regressions included much regard
for student and institutional quality, citing highly significant t-values on variables
measuring student ability and institutional efforts to solicit donations.
In 1996, Baade and Sundberg used a similar version of their model to answer the
question between alumni generosity and athletic success. Fourth Down and Gold to Go?
14
143 public institutions, 112 private universities, and 550 liberal arts schools
15
The proxy was the percentage of student?s receiving student loan aid and will be
discussed later.
15
Assessing the Link between Athletics and Alumni Giving (1996) proves to compensate for
problems associated with much of the earlier work. Their analysis includes many of the
variables left out of the previous literature, focuses specifically on giving to respective
universities, and incorporates schools outside of NCAA Division I athletics.
Data was gathered on 48 private doctorate-granting universities, 94 public
doctorate-granting universities, and 167 liberal arts colleges for the academic years
spanning 1973-1990. This panel study was the most comprehensive than any before it.
However, much of the data gathered was not based on pure theory, but rather on what
was available. They included three athletic measures: (1) a dummy corresponding to
football bowl game appearance, (2) a dummy corresponding to an NCAA tournament
appearance, and (3) respective sports winning percentages.
Their results concluded that most all of the non-athletic variables were significant.
Post-season bowl game appearances were positive and significant at a level greater than
99% in both private and public universities. Using a semi-elasticity method of
conversion, Baade and Sundberg estimated that a bowl game appearance increased
alumni donations at an average public institution by 54% or over $2 million. NCAA
tournament appearance was correlated with increases in contributions only in public
institutions, increasing alumni donations by 35%
In closing their analysis, Baade and Sundberg put focus on athletic programs
advancing to post season play. Winning percentages were never significant in any of
their regressions. Their final suggestion is as follows, ?money spent building the sports
program may in fact reduce alumni giving, if it is at the expense of academic quality.?
(Baade and Sundberg, 802).
16
In summary, the literature discussed above is relatively neutral on the effect of
athletics on alumni support. While more recent studies prove positive relationships
between athletic success and alumni generosity, the majority of articles reveal only
insignificance. The next chapters present several models in which to examine the
relationship between athletics and alumni donations.
17
CHAPTER III
THE DATA
This study utilizes a constructed database of financial, institutional, and athletic
data from schools of higher education during the fiscal year, 2004.
16
Empirical analysis
of the relationship between athletics and giving is constrained by data availability. While
information on the sports success of colleges and universities is widely available, other
institutional factors are often proprietary, unreported, or riddled with measurement error.
These problems present challenges for empirical work in the area.
The majority of the data for this study comes from the Integrated Postsecondary
Education Data System (IPEDS) collected by the Department of Education. IPEDS is a
single, comprehensive system designed to encompass all educational organizations whose
primary purpose is providing postsecondary instruction. IPEDS is developed from a
series of interrelated surveys containing institution-level data on enrollments, graduation
rates, staffing and finances. In several cases universities did not report a complete set of
data to the IPEDS system. These missing data points were collected from another source,
the Peterson?s Guide to Four-Year Colleges. The guide is published annually by the
16
FY2004 reporting dates run from August 2003 to July 2004 for institutions included in
this study.
18
Thomson Company and contains information designed to aid high school graduates in
evaluating different colleges and universities.
17
Data on alumni generosity is obtained from the Council for Aid to Education
(CAE). The CAE is a non-profit group whose initial purpose was improving corporate
support of higher education. More recently, the CAE develops and maintains the most
comprehensive database on alumni information.
18
This information is contained in their
Voluntary Support of Education survey sent out annually to various colleges and
universities. Information in this paper is limited by access restrictions to CAE data. The
CAE requires fees in order to view the entire database; however, selected years and
variables are available on a trial basis.
19
Finally, all information regarding football and
basketball performance is gathered from the NCAA official results book.
20
The resulting
dataset is a cross-section of 658 colleges and universities from across the country.
Previous literature measures alumni giving in several ways. Some of the earliest
work like Admur (1971) and Culip (1965) discussed the effect of athletic success on total
alumni donations. A problem is that the use of total alumni giving does not account for
differences in the number of alumni. An institution with a larger group of alumni, due to
factors like institutional age and enrollment size, will have, perhaps, more alumni
donations. Brooker and Klastoria (1981) were among the first to offer an alternative
measure of alumni giving. As mentioned earlier, these authors measured alumni giving
17
A random sample of schools was selected to compare the data from the two sources. In
all cases, the IPEDS and Peterson?s Guide yielded the same results.
18
The CAE has information regarding the voluntary support of education for over 60
years. Electronic information is available through their VSE Data Minor System.
Subscribers gain access to 10 years worth of data on over 300 variables for thousands of
institutions. Costs associated with the VSE system can be found at www.cae.org.
19
Total alumni contributions is the trial variable used in this study.
20
NCAA official rule book and results is available online at www.ncaa.org.
19
by average contributions from solicited alumni and the per capita gifts to annual funds.
The use of average and per capita measures eliminates the size-of-the-alumni-pool
problem, but other problems remain. Solicited alumni may only account for a small
fraction of the alumni who can and do donate. The problem with using per capita gifts to
annual fund accounts is that most universities have several ?funds? to which alumni can
contribute.
As an alternative measure, Baade and Sundberg (1996) developed a
?contributions per alumni? variable, dividing total giving by the number of alumni of
record. Baade and Sundberg obtained their information from the same CAE database as
used in this thesis. This measure solves the problems with the other measures of giving
previously discussed. However, determining the ?number of alumni? is problematic.
The reported value is subject to the interpretation of the individual school doing the
reporting. In a personal correspondence, Sundberg made the following comment on the
reliability of alumni data:
When we talked to our contact at the Council, he said that the data integrity was
pretty high except in the case of "alumni of record". He said that different
institutions have wildly different attitudes toward that. In some cases, alumni of
record include virtually all non-deceased alums, while in others it consists of only
those alumni for whom the institution has current addresses. (Sundberg)
This quote suggests the possibility of measurement error. Rhoads and Gerking (2000)
also note the high degree of measurement error in the CAE data. Their attempt to scale
total alumni contributions involved dividing through by the respective school?s
enrollment. Their ?average gift per student? variable solved the institutional size
problem. Rhoads and Gerking also included an age of the institution variable in their
regressions as to control for the effects of age.
20
The dependent variable chosen for use in this thesis is determined by a
comparison of four different measures based on previous research. Table I displays the
four variables, a definition of each, and their respective descriptive statistics. In the
following paragraphs each of these four variables is discussed in turn.
AlumTotal is the total alumni contributions made to an institution by alumni in the
2004 fiscal year. This measure includes a variety of giving categories and is presented in
total dollars.
21
Due to problems with this measure previously discussed, AlumTotal is
included for purposes of comparison.
AlumPS is total contributions divided by the enrollment of the institution and so
controls for differences in the university size.
22
AlumPS is calculated using total
contributions (AlumTotal) divided by total enrollment in the fall of 2003.
Two other measures, AlumPA and AlumPA2 involve approximations for the actual
number of alumni at an institution. The goal is to eliminate or minimize the error in
reported alumni numbers discussed previously. AlumPA is defined to be
(1)
)lim)(25.0(
1000
ii
i
i
AgeEnroll
AlumTotal
AlumPA
?
=
where total alumni contributions is divided by a proxy for the number of alumni. The
quarter enrollment is meant to represent the size of the graduating class (future alumni)
and, multiplying by the school?s age, yields an estimate of the number of alumni. The
school?s age, Age, is restricted to be less than or equal to 100 years. This is done because
alumni graduating 100 years ago would no longer be living to contribute. This
21
Giving categories include, but are not limited to: annual fund, capital campaign, and
earmarked gifts (towards academics or athletics). However, only donations from alumni
are counted.
22
See Rhoads & Gerking (2000)
21
constructed variable, AlumPA, is not without problems. For example, the 0.25 times
enrollment is an overestimate of actual graduation rates; it would require a university
having constant enrollments over time and a 100 percent retention and graduation rate.
AlumPA2 uses another measure of total alumni. This variable is similar to
AlumPA except the upper limit on institutional age of 100 years is removed and actual
age is used. Essentially this is a measure of the number of alumni in the history of the
school.
The independent variables used to estimate differences in levels of alumni
generosity are based upon economic theory and previous literature. Consumer demand
theory suggests that income, tax incentives, and tastes should explain differences in
giving behavior. The explanatory variables chosen in this thesis control for these three
factors.
23
Institutional characteristics are also included due to the specific nature of
alumni contributions. Data on all explanatory variables is based on the 2004 fiscal year,
or the corresponding 2003-2004 academic year.
Most universities keep few, if any, records of alumni income. Even if this data
were collected it is not likely to be publicly available. To estimate income effects,
current student information is used. Previous research uses this information for two
reasons: to examine the future donation potential of a student and to control for
differences in the family wealth. The next six variables adopt this approach with a
variety of improvements.
23
AvgSAT, Tuition, %Min, %Fem, %RecLoan, and AvgLoan describe potential income,
Tax measures tax liability, StudExp explain the differences in taste, while Age, Enroll,
and Private determine institutional characteristics.
22
The first two variables are based off student demographics. Information on the
minorities and females as a percentage of total enrollments for the 2003-2004 school year
are included in the regressions. Data on minority enrollments is determined by dividing
Caucasian enrollment by total enrollment then subtracting from 1 (%Min). This measure
includes African American, Asian American, Native American, Hispanic, Middle Eastern
and those not reporting their ethnic status. Female enrollments (%Fem) are similarly
determined; dividing female enrollment by total enrollment.
These characteristics aid in determining the giving ability of respective alumni.
Previous research has shown that minorities and women begin employment at lower
starting salaries than their majority (male) counterparts.
24
For that reason, it is likely to
assume that these individuals will leave their alma mater with, initially, a smaller
capacity to donate. The consequence is that higher percentages of minority and female
enrollments (i.e. alumni) are associated with less giving.
Baade and Sundberg (1996) introduced the ?family wealth? idea of alumni
generosity. This is the notion that wealth passing from generation to generation is
available for future alumni giving. Simply put, wealthier families have greater capacity
to contribute. Baade and Sundberg developed two proxy measures for ?family wealth.?
The first is the percentage of students at an institution receiving financial aid. Students
who accept financial aid likely do not have the wealth available to finance college. A
higher percentage of students on financial aid implies poorer families. Therefore, one
expects an inverse relationship between the percentage of students on financial aid and
alumni generosity. This thesis includes a similar but improved measure of family wealth.
24
See Oaxaca (1978)
23
General financial aid includes measures like loans, grants, and scholarships. The former
two components, grants and scholarships, are often performance-based stipends and do
not necessarily reflect financial need. Students excelling academically receive this
assistance, but that does not imply that these students lack the capacity to pay for college.
In this thesis the percentage of students receiving loans (%RecLoan) will be used as a
measure of family wealth. This is expected to be inversely related to alumni
contributions.
The second gauge of family wealth given by Baade and Sundberg (1996) is the
average amount of financial aid per student. Like before, higher values denote lower
overall wealth. Using the argument above, this thesis employs the average amount of
student loan aid (AvgLoan). AvgLoan is expected to vary inversely with alumni
contributions. All values for %RecLoan and AvgLoan represent the 2003-2004 academic
year.
Another proxy for family wealth is tuition costs. Tuition is the total of tuition and
fees required for the 2003-2004 academic year.
25
Tuition costs indicate the level of
wealth needed to avoid use of financial aid. Again, wealthier families have enough
money to cover the cost of tuition and should have a higher capacity to donate.
Accordingly, tuition should be positively related to alumni giving. In fact, there may be a
second avenue for the effect of tuition on giving. Tuition costs are often positively
related to quality of education. Better education (higher cost) leads to higher-paying jobs
and, perhaps, an increased capacity to donate.
25
For public universities, Tuition is the instate value for all cost and fees.
24
Most of the previous literature includes some measure of student quality as
another proxy for alumni income.
26
Brighter students often get higher paying jobs and
enjoy an increased capacity for giving. Standardized test scores (average SAT or ACT
equivalent) found in admission reports are used here (AvgSAT). AvgSAT is expected to
be positively related to alumni giving. In cases where schools reported only ACT
scores, the values were transformed into an SAT equivalent using a conversion guide
produced by the College Board.
27
All values for AvgSAT refer to the incoming class for
fall 2003.
Previous articles include tax-related variables to account for deductions on
charitable contributions. Differences in tax incentives are incorporated by including an
effective marginal tax rate using methodology developed in Long (2000). The measure is
a summation of federal tax liability on the average adjusted gross income of the state in
which the institution is located and the highest marginal state tax, if applicable. Itemized
and federal tax deductions on state returns are included as well. The tax rate is defined
as:
(2)
)1(
)(
s
sss
fst
fsdfstsdf
Tax
?
??+
=
where f is the federal marginal tax rate faced by the average adjusted gross income, s is
the highest marginal state tax rate, t
s
is a value equal to ?1? if federal taxes can be
deducted from state returns and ?0? if otherwise, and d
s
equals ?1? if the state allows
itemized deductions on state returns and ?0? otherwise
26
See Baade & Sundberg (1995,1996) or Rhoads & Gerking (2000)
27
College Board is a non-profit group providing information to students entering post-
secondary education.
Their equivalency charts can be found at www.collegeboard.com
25
Many times an individual?s experience at a particular university provides the
foundation or ?tastes? for giving behavior. An alumnus who enjoyed his or her collegiate
years will be more inclined to make donations to that institution. However, the quality of
one?s college experience is a difficult variable to measure. Students pay a price for
college, mainly tuition. In return, students receive an education and an experience. This
experience includes things like extracurricular activities and academic support.
University expenditures on students are used as a proxy for ?experience.? The more a
university spends towards students, the richer the student?s time will be. To construct
this proxy, data was gathered on institutional budgets. For each university, three main
categories of spending for the FY2004 (instruction, academic support, and student
services) were summed and divided by total enrollment to yield the expenditure per
student variable (StudExp). Baade & Sundberg (1995, 1996) use two similar variations
on this measure: instructional expenditure per student and scholarship expenditure per
student. StudExp is chosen for use in this thesis because it represents a direct measure of
spending on students. There is assumed to be a positive association between StudExp and
alumni generosity.
Three institutional characteristics are included as explanatory variables. The first
two, Enroll and Age, attempt to measure the relative size of the alumni pool. Enroll is the
total full-time enrollment for the 2003-2004 academic year. More students at a university
represent a larger potential alumni body. Enroll is expected to be positively associated
with giving. Age is defined as the current year (2003) minus the founding date of the
institution. Older schools may not only have more alumni, but also a history of giving.
Age is also assumed to be positively related to donations. Lastly, a dummy variable is
26
included to allow for differences between public and private institutions. Private schools
rely on support primarily from tuition dollars and alumni giving. Public schools have
assistance from state appropriations to aid them in funding. In order to account for this
difference, Private is defined to equal to ?1? for private institutions and ?0? for public.
Information on school efforts to solicit alumni may also have bearing in the
model. However, relevant data on university campaigns is not consolidated and often
inconsistently reported. Most colleges have alumni associations or branches of the
university administration that deal with alumni.
Table II presents summary definitions for each of the explanatory variables and
the expected effect on alumni generosity. Table III shows the descriptive statistics.
27
CHAPTER IV:
DETERMINING THE EFFECT ATHLETIC PROGRAMS
The objective of this thesis is to determine the relationship between collegiate
athletics and alumni generosity. Past research includes only athletic ?success? measures.
These articles explored the relationship between athletic success and alumni giving. This
thesis begins with an investigation into the relationship between alumni giving and the
presence of an athletic program, successful or not.
To examine this correlation, one could simply calculate the mean difference in
giving between schools with and without athletic programs. For example, Table IV
presents the average values of several measures of giving. Means are calculated both for
schools with and without NCAA football programs. As the table shows, an average
university participating in football will receive over $10 million in alumni contributions.
This is over $8 million more than a school without a football program. However, this
difference does not account for variation in other factors that might influence giving in
football and non-football schools.
In order to control for these other differences, regression analysis will be used.
The variables discussed in the previous chapter provide the basis for the regression model
of alumni generosity. Two variables are introduced to account for any effect of
intercollegiate athletics. The first is Football. A dummy variable is constructed to equal
?0? for schools not participating in NCAA football and ?1? for those participating
28
schools.
28
Similarly, Basketball is a dummy variable equal to ?0? for those schools not
involved in NCAA Basketball and equal to ?1? for schools that are. Football and
basketball are the only two sports considered in this analysis because they are
traditionally revenue-generating sports and receive the majority of the media exposure.
Much of the previous literature in this area employs a double log model. This
model allows the authors to control for large degree of variability in giving and potential
heteroskedasticity. Estimated values in log-linear models are easily interpreted as
elasticities. However, variables like Private, Football and Basketball are dummy values
and cannot undergo the transformation. Expressed in linear form, the percent change
interpretation no longer applies. Many authors use a semi-elasticity transformation, but
this approach produces biased estimates.
29
In addition, the interpretation of the Blinder
decomposition results in the next chapter is much easier in a linear model. For these
reasons the linear functional form is chosen for use in the empirical work.
30
The OLS regression results from the estimation of four linear specifications for
alumni generosity model are presented in Table V. The dependent variable is different in
each model as discussed in the previous section: AlumTotal, AlumPS, AlumPA and
28
It is worth noting that schools participating in NCAA football and basketball constitute
all competitive divisions. Further, schools with a ?0? value (i.e. schools not participating
in NCAA) do not necessarily lack athletic teams; these institutions simply do not
participate in the NCAA.
29
The semi-elasticity calculation uses an exponential function to convert a logarithmic
value. The exponential transformation is a non-linear function and computes an biased
estimate. See Wooldridge (2003).
30
Several other reasons for dismissing log transformation are cited. Data points with a
zero or negative value cannot undergo the natural log transformation. Later analysis
incorporates winning percentages, of which several cite 0 values. Also, %RecLoan,
%Min, and %Fem are measured in percentages. Their log estimates would be interpreted
as point percentages (i.e. a 1 percent change in a percent) and may prove difficult to
translate.
29
AlumPA2, respectively. Any explanatory variable involved in the calculation of a
dependent variable is omitted from the regression model to eliminate spurious correlation
or enhanced explanatory power.
Models I uses AlumTotal as the dependent variable. Regressed against a set of 13
independent variables, the model reports the highest R
2
value of 0.49. Model I has seven
significant coefficients at a 95 percent level or better (Enroll, Age, AvgSAT, Tuition,
%Min, StudExp, and Private). Tuition and %Fem, reporting parameter values of -360.92
and 3.45 x 10
6
respectively, are the only variables inconsistent with expectations. Both
sports related dummy variables (Football and Basketball) have t-ratios below critical
levels and are insignificant.
Model II utilizes AlumPS as the dependent variable while dropping Enroll from
the right hand side. The model has eight significant coefficients with an R
2
value of 0.37.
While Tuition and Private drop from statistical importance, measures of financial aid
(%RecLoan and AvgLoan) and female enrollments report significant impact. Football
becomes significant (only at a 90 percent level), showing a positive coefficient of 351.57.
Basketball remains inconclusive.
Using AlumPA as the dependent variable, the Model III explains over a third of
the variation in alumni giving (reporting an R
2
value of 0.35) with seven significant
coefficients.
31
Both Enroll and Age are dropped from the equation. Generally all
variables follow their expected signs with exception to %Fem. Football increases its
significance to a 95 percent level, while the t-ratio on Basketball continues drop.
31
AvgSAT, %RecLoan, AvgLoan, %Min, %Fem, StudExp, and Football are all significant
at a 90% level or better (the intercept is not included as an independent variable, though it
is significant).
30
Model IV employs AlumPA2 as the dependent variable. It reports the lowest R
2
value of 0.31 and the smallest number of significant coefficients, four. %RecLoan, %Min
and Football drop below statistically significant levels and Basketball continues to
remain insignificant.
Model III is the best model fit for this analysis. While Model II reports a higher
R
2
(0.37), Model III is chosen on the basis of stronger t-ratios. The majority of
significant coefficients in Model III report higher t-ratios than Model II, with exception to
%RecLoan, AvgLoan, and StudExp (though the difference is minute). Of particular
importance is the increased explanatory power of the football dummy variable. The t-
ratio increases from 1.80 to 2.35 when moving from Model II to III. The additional
statistical strength of Football and other variables outweighs the 0.02 loss in R
2
value.
The chosen regression equation is presented below,
(3)
ii
iiiiii
iiiii
Basketball
FootballivateTaxStudExpFemMin
AvgLoancLoanTuitionAvgSATAlumPA
??
??????
?????
++
++++++
++++=
11
1098765
43210
Pr%%
Re%
where i represents the individual cross-section and ? is a disturbance term. Implications
of the coefficient values will be discussed later.
Several diagnostic tests are performed prior to any detailed discussion of the
results. While its presence does not lead to a violation of OLS assumptions,
multicollinearity may produce imprecise estimates and wider confidence intervals. Two
methods are used to detect the presence of collinear relationships among the independent
variables. The first is the calculation of the matrix of Pearson correlation coefficients
presented in Table VI. This table shows a very high correlation between Tuition and
Private (0.87). The correlation makes sense given that the mean tuition at a private
31
school is over $15,000 more than the mean tuition at a public institution. Further
investigation into the presence of multicollinearity is achieved by computing the
Variance Inflation Factors (VIF) for each explanatory variable (see Table VII). These
measures are scaled versions of the multiple correlation coefficients between each
independent variable and the remainder of the independent variables. All VIF values are
below 10, ruling out multicollinearity.
32
Further diagnostic testing was conducted to investigate the presence of
heteroscedasticity, which is likely to be present given that dependent variable is related to
expenditures. Heteroscedasticity leads to inefficient estimates, underestimated standard
errors, and overestimated t-ratios. White?s test is used to investigate the presence of
heteroscedasticity. Residual and predicted values are obtained from the original OLS
regression equation. Residuals values are squared and used as the dependent variable in
two regression models including, in turn, the predicted values and the square of the
predicted values as independent variables. From these two regressions, a statistic of
22.15 is obtained which is less than the critical ?2 value, indicating that the null
hypothesis of a homoskedastic error term cannot be rejected.
33
In addition, a Breusch-
Pagan test is also administered. Here, the squared residuals from the linear equation are
regressed on the original explanatory variables. The calculated LM statistic of 672.05 is
significantly above the critical ?2 value; denoting the presence of heteroscedasticity.
32
It is generally accepted that a value greater than 10 is an indication of potential
multicollinearity. For reference, see Neter, Wasserman, and Kunter (1990).
33
Critical 2 value is 68.67 at 95% significance
32
The tests for heteroscedasticity give conflicting results. In order to be cautious,
White?s correction is used to produce heteroscedasticity-robust standard errors and
consistent t-ratios. Table VIII reports the corrected regression output.
As mentioned before, the model contains seven significant coefficients at the 95%
level or better. Most have the expected sign suggested in the conceptual model. The
directions of the effects of several of the included student characteristics are consistent
with their hypothesized effects. Higher SAT scores are associated with higher alumni
giving, while higher minority enrollments and higher average student loans reduce
giving. All significant institutional factors (StudExp and Private) have positive
associations with contributions. Universities spending more on their students experience
higher donations, possibly due to the enhanced experience felt by alumni. Private
schools also experience more giving. The results indicate that the average private school
enjoys $4.6 million more in alumni giving than its public counterpart. Tuition and
%RecLoan both have the expected sign, but fall just short of statistical significance. The
coefficients of Tax and Basketball have very small t-ratios.
One surprising result is the positive effect of higher female enrollment. Past
research suggests that female enrollments are typically negatively related to alumni
giving. The positive coefficient value of 91.30 on %Fem suggests otherwise. A one
percent increase in female enrollments would increase alumni generosity per alum by
roughly $0.91 which, in a university of average size, would lead to an increase in giving
of $152,541.
34
34
An average institution is based off the mean values of the sample. It would be 122
years old, enroll 7501 students and have 167,957 potential alumni.
33
Schools investing in intercollegiate football see substantially larger alumni
contributions than those that do not. The average university competing in NCAA football
could enjoy over $3.3 million more per year in alumni generosity than schools not
participating in football. The argument becomes more convincing when considering the
remaining significant coefficients. To match the $3.3 million increase in giving, an
average university would have to increase its female enrollment by 22 percent, increase
student expenditures by over $45 million or raise the average SAT score of an entering
freshman 143 points, roughly 13 percent.
Questions arose to the relationship of football and the explanatory variables in the
giving model. Several of the independent variables may factor into whether or not a
school might have a football program. To determine any level of endogeneity in
Football, a Hausman test was constructed. An equation for football participation was
developed using Football as the dependent variable and the remaining independent
variables from the original regression equation (3). Also included on the explanatory side
are a total revenue per student variable and a dummy variable indicating participation in
the NCAA.
35
After undergoing OLS estimation, the residual values were saved and
placed inside the alumni giving equation (3). The results of the second regression
showed a t-ratio on the residual values of 1.51. Insignificant at a 90 percent level, this
demonstrates no endogeneity associated between alumni generosity and football
participation.
Clearly collegiate football programs have a strong bearing on alumni giving.
Football games provide a level of exposure to respective alumni not otherwise given by a
35
RevPS was constructed using total revenue divided by total enrollment. NCAA, the
dummy variable, indicates membership in the NCAA no matter what athletic sport.
34
university. Television and print media coverage act as school advertisements, while
game attendance entices alumni to return to their alma mater. Past research often refers
to this phenomenon as a ?warm glow? effect, whereby alumni feel connected to their old
school through football competition.
36
Whatever the reasoning, college football programs
do provide some spill over benefit relating to increased alumni giving.
The next section will present separate regressions for football and nonfootball
models and a decomposition of the difference between the two. For several reasons, the
dummy variable on basketball will be dropped and focus centered on football. Basketball
demonstrates low t-ratios in all the regression results presented above and is continually
deemed statistically insignificant. Further, the effect of basketball is captured by a
portion of the football dummy. All schools within the data set that participate in NCAA
football also participate in NCAA basketball. Future relationships involving basketball
will be presented in a later chapter.
36
See Turner, Meserve, and Bowen (2001)
35
CHAPTER V
BLINDER DECOMPOSITION
As indicated in the previous section, the presence of a college football program is
associated with a sizeable increase in alumni revenues. However, the significance of the
dummy variable, Football, means only that the intercept for football schools can change.
The coefficients of all other independent variables are forced to be the same for football
and non-football schools. This may not be true. Previous literature fails to address the
issue of whether the alumni giving equation differs between schools with and without
football programs. In testing this hypothesis, two new regressions models are compared:
the original model minus the football dummy variable (essentially equation (3)), and a
fully interacted model using Football as the integrating factor. An F-test indicated the
null hypothesis that all regression coefficients are equal between football and non-
football schools can be rejected.
37
This finding verifies separate regressions for football and nonfootball schools are
appropriate. The OLS results of these two models are reported in columns 2 and 3 of
Table IX. The non-football model explains a greater portion of the variation in alumni
giving (R
2
of 0.45 as opposed to 0.32 for the football model). Each model contains five
significant coefficients. Coefficients on AvgSAT, AvgLoan, and StudExp are significant
in both models. Obvious differences in explanatory significance are found in the
37
The calculated F-statistic was 2.73 with a critical value of 1.86 and 2.38 for =0.05 and
=0.01 respectively.
36
coefficients of %Min, %Fem, Tax and Private. In the football regression, the remaining
significant variables are minority enrollments and the private/public status of an
institution. In the non-football regression, the two other significant variables are percent
female enrollment and the tax price. Tax has a coefficient value of -316.87 at a
significance greater than 95 percent in the non-football sample.
Given the two regressions, it is possible to decompose the average difference in
giving into two distinct components. The first is the difference resulting from the school
characteristics such as average SAT, tuition costs, and various student demographics.
The second is the coefficient effect or the differing marginal effect of changes in the
independent variables on the alumni generosity equation. This paper employs an
Oaxaca-Blinder decomposition, often found in labor economics as a means for
determining discrimination.
38
The mean alumni generosity per alumni value for football ( fG ) and non-football
(
nG
) schools can be written as:
(5) )( F
F
F XG ?=
and
(6) )( N
N
N XG ?=
respectively, where ? denotes the vector of regression coefficients and X is the vector of
the mean values of the explanatory variables. Using the equations above, the difference
in alumni generosity between football and non-football schools can be expressed as,
38
For an example, see Oaxaca (1973) or Blinder (1973). For specific methodology used
here, see Long & Caudill (1992).
37
(7) )()( N
N
F
F
NF XXGG ?? ?=?
After some simple manipulations, the ride hand side of (7) now becomes,
(8) )]()([)]()([ N
N
N
F
N
F
F
F
XXXX ???? ?+?
The first difference shows the disparity in the mean values evaluated using the
coefficients for football schools. This can be viewed as the difference attributed to
differing values of school characteristics. If found to be positive (negative), then the
disparities in school characteristics favor football (non-football) institutions. The second
portion of the equation shows the differences in coefficient values of football and non-
football universities evaluated at the non-football means. This accounts for any
discrepancy in the way the characteristics of football and nonfootball schools affect
generosity.
39
These parameter differences are the unexplained variation between the
schools and could be attributed to football participation, omitted variables, or possible
misspecification in the alumni generosity equation.
40
Table X shows that the proportion of the difference in alumni generosity between
football and non-football institutions attributed to disparities in school characteristics is
only 19.92 percent, leaving 80.08 percent of the variation unexplained. To appreciate the
results, the following example is presented.
An average NCAA football-participating university has a mean AlumPA value of
$71.56, while schools without football have a mean value of $44.03. Multiplying each
39
If there was no difference in the model treatment of football and non-football schools,
the value of )]()([ N
N
N
F
XX ?? ? would be zero.
40
The equation can be equally evaluated as )]()([)]()([ F
N
F
F
N
N
F
N
XXXX ???? ?+? .
The results are consistent with the above mentioned analysis and can be found in Table
XI.
38
by their average number of alumni, the total alumni contribution becomes $15,328,096
for football schools and $4,440,449 for non-football schools. Differences in mean values
explain only $877,554 of the $10,887,647 variation in generosity, leaving over $10
million depending on the perceptions of football at these schools.
Table X presents a partition of the individual influences of the explanatory
variables in the decomposition. Individual effects are reported as percentages with a
weighted t-ratio presented below.
41
Five variables are associated with statistically
significant effects on the giving levels between the two groups of universities. Average
SAT scores, student expenditures, and marginal tax rates all show positive percent
values, indicating a disproportional preference for football schools. These factors
account for 71 percent of the positive variation in the observed generosity difference.
While the average loan amount is positive and significant, the variable controls for only 3
percent of the difference and hence relatively inconsequential. %Fem reports a
significant negative percentage. This indicates that higher female enrollments are less
favorable for football schools and contribute more towards universities without football.
Table X also demonstrates that female enrollments, tax rates, and private status all
have significant weight in the unexplained variation between football and non-football
institutions. As mentioned earlier, this unexplained difference can be attributed to the
university participating in NCAA football. Again, %Fem is associated with a negative
percentage and can be interpreted as increasing alumni generosity at universities without
football programs. Tax and Private both show positive percent values and hence a
41
Weighted t-values are constructed using the respective standard error of the football
(non-football) schools and the relative percentage make-up of the sample. For complete
methodology, see Long and Caudill (1992).
39
favorable treatment towards increasing alumni contributions at football schools. These
findings lead to the following conclusions. Women seem to be less inclined to donate to
an alma mater that invests in football. Female students (potential alumni) may be more
concerned with the academic nature of higher education and may find football a needless
distraction. However, tax rates and private status account for increases in alumni giving
at football schools. Football school preference of the tax rate (i.e. the incentive to donate
for income deductions) could indicate higher incomes within those schools. The positive
effect of Private is consistent with the earlier argument that private schools are better
schools leading to better jobs
A tale of two schools: Auburn University and Emory. In 2003, Auburn
University had an enrollment of 21,236, an estimated alumni base of 531,775, and
received $18,529,853 in alumni contributions. During that time, Auburn participated in
NCAA Division I-A football. However, what would the contributions have been if
Auburn had not participated in football? Using the coefficients from the non-football
model in Table IX, along with the Auburn characteristics for 2003, a prediction can be
made. Without football, the school could expect to receive $9,287,267, or about half of
the original amount. On the other hand, consider Emory University, a school without a
football team. With an enrollment of 11,113 and alumni contributions of $7,425,514, the
same calculation can be made by inserting Emory?s characteristics into the football
regression. With football, Emory might expect to see as much as $50 million more in
alumni contributions.
40
CHAPTER VI
ESTIMATION OF ATHLETIC SUCCESS
As was evident from the literature review, most of the research involving giving
and athletics has focused on the effect of athletic success. Every study uses athletic
performance measures to incorporate the effect of athletics. These include winning
percentages, postseason appearances, and national rankings. The issue is examined in
this thesis as well.
The sample used here is a subset of the sample from the previous chapters.
Because athletic success measures are introduced, schools without athletic programs must
be omitted from the sample. The resulting dataset is a cross-section of 357 institutions
participating in both NCAA football and basketball. Again, all variables are taken from
the 2004 fiscal year and corresponding 2003-2004 academic year. Athletic success
values are collected from the previous three sports seasons, beginning fall 2002 and
ending spring 2004.
Athletic success is measured using two types of variables.
42
The first are the
winning percentages of a university?s football and basketball programs. These values are
calculated by dividing the total number of wins by the total number of games played in a
season.
43
WPFB
-2
, WPFB
-1
, and WPFB correspond to the winning percentages from the
42
Descriptive statistics of the athletic success variables can be found in Table XIII.
43
Total number of wins and games played includes any postseason play.
41
2001, 2002, and 2003 football seasons.
44
WPBB
-2
, WPBB
-1
, and WPBB are the respective
winning percentages for the 2001, 2002, and 2003 basketball season. The second set of
performance measures is a series of dummy variables indicating appearances in the
postseason. All NCAA divisions offer some postseason opportunities for teams meeting
high performance criteria; most notably the Bowl Championship Series and NCAA
Basketball Tournament in Division I athletics. FBPS
-2
, FBPS
-1
, and FBPS take on a ?1?
value for appearances in the 2001, 2002, and 2003 postseasons. BBPS
-2
, BBPS
-1
, and
BBPS are the basketball counterparts.
The lagged values of the success measures are included to control for time
sensitive donations towards universities. Donors who are motivated by athletic
performance may not have time to contribute prior to the start of the next fiscal year. For
example, consider the NCAA Basketball tournament, which starts in March and finishes
around mid-April. Consideration was also given to include differenced values for the
athletic success variables. It was felt that some giving from athletic performance is based
on relative success, i.e. how well a team did from the previous year. Ultimately, the
difference variables were not retained. The coefficient on a differenced variable would
essentially represent the difference between the current and lagged values (already
included in the regression). Also, there is no definitive way to construct difference values
on the postseason dummy variables.
Previous research recognizes the competitive differences between division levels
in college athletics. Sigelman and Bookheimer (1983) focused their entire study on a
collection of ?big time? athletic programs, that is, those in Division I-A. Rhoads and
44
Athletic seasons (postseasons) run from July to June, i.e. the 2001 season ran from July
2001 to June 2002
42
Gerking (2000) note that Division I schools make long-term, higher cost commitments to
high-profile athletic programs. In fact, the sheer volume of media exposure of the Bowl
Championship Series and March Madness is reason enough to separate the sample into
divisions.
Table XIV presents the mean values of the variables used in this study, sorted by
athletic division. The mean values are surprising. For example, an average Division III
university receives almost $30 more per alumnus than an average Division I school.
However, when looking at total alumni giving, the average DI institution receives $20
million more in generosity than an average DIII school. Again, this analysis is without
any control for other factors affecting donations. For this reason, OLS regression is used
to determine the specific effects of athletic success on giving.
The model used here is nearly identical to the model used in the preceding
analysis with athletic success variables included in place of participation measures. The
model is:
(9)
iiii
iiiii
iiiii
iiiii
iiii
BBPSBBPSBBPS
WPBBWPBBWPBBFBPSFBPS
FBPSWPFBWPFBWPFBivate
TaxStudExpFemMinAvgLoan
cLoanTuitionAvgSATAlumPA
????
?????
?????
?????
????
++++
+++++
+++++
+++++
+++=
??
???
???
21120219
1811721615114
213121112109
87654
3210
Pr
%%
Re%
Table XIV presents the results of four regressions using the model above. Each column
in the table corresponds to a different competitive division with column 1 being the total
sample. Diagnostic testing revealed no major problem with multicollinearity and the t-
vales reported have undergone White?s correction.
43
When analyzing all divisions together, the estimated model has four significant
coefficients (AvgSAT, AvgLoan, StudExp, and Private) and an R
2
of 0.34. However,
there are no significant coefficients associated with any of the athletic success measures.
In column 2, looking only at Division I schools, the coefficient of Private is no longer
significant and the R
2
remains at 0.34. The current winning percentage and a bowl game
appearance last season both negatively affect giving at a 90 percent level of significance.
The results for Division II schools are given in column 3. The model R
2
is 0.65 and the
statistically significant non-athletic variables include Tuition, %Min, %Fem, Tax, and
Private. Winning percentages in both basketball and football from the 2001 season are
statistically significant; football success has a negative effect and basketball success has a
positive effect. The results for Division III are given in column 4 of the table. The model
has an R
2
of 0.43 and contains three statistically significant variables (AvgLoan, %Fem,
and StudExp); however none of the athletic performance variables are statistically
significant.
Because each of these regressions contains a group of athletic performance
variables, F-tests are conducted to determine whether ?athletic performance? helps
explain differences in alumni giving. Table XV reports F-values associated with testing
the joint significance of all athletic success variables in each model. The computed F-
values are 0.72, 0.93, 0.78 and 0.39 for the total, DI, DII, and DIII regression models,
respectively. All of the values fall below their critical levels and the null hypothesis that
all of the coefficients of the athletic success are zero cannot be rejected. Athletic
performance does not seem to matter when explaining differences in alumni giving.
44
CHAPTER VII
CONCLUSION
The case for college athletics is, and will continue to be, scrutinized by most
everyone in and around academia. Though recording overall revenues in the billions of
dollars, roughly 60 percent of university athletic programs operate with net losses. Such
financial results have spurred researchers to investigate other effects of college sports.
This thesis has examined the role intercollegiate athletics plays in encouraging
contributions from alumni. A cross-section of 657 schools for the FY2004 provides the
basis for this study. The inclusion of schools without athletic programs in this dataset
allows for a more thorough investigation than previous literature has offered.
The initial analysis explored the effect of athletic participation by a university; a
question also never considered in previous literature. A constructed model of alumni
generosity yielded several significant student, alumni and institutional characteristics.
SAT scores and female enrollments are both directly related to giving, while the average
student loan amount and minority percentages affect inversely. University characteristics
like student expenditures and private/public status have positive connections with
educational contributions.
It can be said that participation in NCAA football has a positive effect on alumni
donations. School having such a program can expect roughly $3.3 million more in
45
educational contributions than a university choosing not to participate. However, results
offer no significant relationship between basketball programs and levels of giving.
Decomposition reveals over 80 percent of the differences between alumni
contributions at football and non-football schools are determined by the sheer presence of
a football team. Alumni treat schools with football programs more favorably by
awarding them with higher donations. Reasons for this effect could stem from media
exposure, feelings of connection from alumni, and a variety of other factors. It was also
shown that women have a negative preference for football while tax rates and private
status encourages donations towards football-participating institutions.
Further analysis was conducted to determine if the relative performance of
athletic programs have any affect on giving. Winning percentages and post-season play
variables were included into the alumni generosity model. The results were inconclusive.
While regressions by competitive division yielded a few significance performance
measures, overall model testing canceled these minute effects. Participation is what
matters, success is inconsequential.
Although the results of this thesis offer insights into the relationship between
athletics and alumni giving, the scope of their interpretation is limited. Data availability
and quality are two issues that this paper, like many others before, find problematic.
Most of the data problems are with the alumni giving information. Alumni data is
inconsistently reported and filled with measurement error. Future studies may try more
direct measures of alumni characteristics, if available. Also, the cross-sectional data used
in this thesis limits any investigation of time-sensitive variation in alumni generosity.
Panel data over several years may better describe schools with traditions of giving or help
46
account for yearly differences in institutional characteristics. Finally, this author
recommends that continued investigations into the relation of alumni giving and athletics
focus on a case-by-case basis. Universities and other institutions of higher education
differ on many levels. Controlling for these differences leads to models that may
misrepresent the characteristics of an individual school. Case studies incorporate these
school-specific variations and could reveal common results among differing institutions.
All in all, university officials wanting to increase educational contributions may
find the results of this thesis beneficial. However, specific policy applications are
imprudent. While there are elevated levels of giving associated with football programs, a
complete cost-benefit analysis should be conducted before any investments in athletics
are undertaken. These findings relate only to alumni contributions and are proof of
spillover benefits in intercollegiate athletics.
47
REFERENCES
Amdur, Neil, 1971. The Fifth Down: Democracy and the Football Revolution. New
York, NY: Coward, McCann and Geoghagen.
Baade, Robert A. and Jeffery O. Sundberg, 1996. ?What Determines Alumni
Generosity,? Economics of Education Review, 15 (1):75-81.
---------, 1996. ?Fourth Down and Gold to Go? Assessing the Link between Athletics and
Alumni Giving,? Social Science Quarterly, 77 (4):75-81.
Becker, G.S. 1974. ?A Theory of Social Interactions.? Journal of Political Economy,
82:1063-1093.
Blinder, A. S., 1973. ?Wage Discrimination: Reduced Form and Structural Estimates,?
Journal of Human Resources, 8: 436?455.
Brooker, George, and T. D. Klastorin, 1981. ?To the Victor Belong the Spoils? College
Alumni Giving.? Social Science Quarterly, 62:744-750.
Caudill, Steven B. and James E. Long, 1992. ?Racial Differences in Homeownership and
Housing Wealth, 1970-1986,? Economic Inquiry, 30:80-100.
Coughlin, Cletus C., and O. Homer Erekson, 1984. ?An Examination of Contributions to
Support Intercollegiate Athletics, ? Southern Economic Journal, 51 (July):180-
195.
Council for Aid to Education. VSE Data Miner System. 2004 Fiscal Year, www.cae.org.
Cutlip, Scott M, 1965. Fund Raising in the United States: Its Role in America?s
Philanthropy. New Brunswick, N.J.: Rutgers University Press.
Feldstein, Martin, 1975. ?The Income Tax and Charitable Contributions: Part II ? The
Impact on Religious, Educational and Other Organizations,? National Tax
Journal, 28:209-226.
Frank, Robert H. and Phillip J. Cook, 1996. Winner-Take-All Society: Why the Few at the
Top Get So Much More Than the Rest of Us. New York, NY: Penguin Group.
48
Goff, B., 2000. ?Effects of University Athletics on the University: A Review and
Extension of Empirical Assessments,? Journal of Sport Management, 14: 85?104.
Greene, William H., 1997. Econometric Analysis. New York, NY: Pearson Education
Inc.
Griffiths, William E., R. Carter Hill and George C. Judge, 1993. Learning and Practicing
Econometrics. New York, NY: John Wiley and Sons, Inc.
Keating, B, 1981. ?United Way Contributions: Anomalous Philanthropy.? Quarterly
Review of Economics and Business, 21:114-119.
Long, James E., 2000. ?Omitted-Variables Bias When Using State Tax Rates to Estimate
the Tax Price Effect on Itemized Deductions,? Public Finance Review, 28:120-
133.
Mixon, F. G., Jr.,1995. Athletics Versus Academics? Rejoining the Evidence from SAT
Scores. Education Economics, 3:277?283.
--------- and Len J. Trevino, 2005. From Kickoff to Commencement: the Positive Role of
Intercollegiate Athletics in Higher Education,? Economics of Education Review,
24:97-102.
Murphy, R.G., and G.A. Trandel, 1994. ?The Relation Between a University?s Football
Record and the Size of Its Applicant Pool,? Economics of Education Review, 13
(September): 265-270.
National Collegiate Athletic Association. 2001-2003. The Official NCAA Men?s
Basketball Records Book. NCAA: www.ncaa.org.
----------. 2001-2003. The Official NCAA Football Records Book. NCAA: www.ncaa.org.
Okunade, A. A., P. V. Wunnava and R. Walsh, Jr., 1994. ?Charitable Giving of Alumni:
Micro-data Evidence from a Large Public University,? American Journal of
Economics and Sociology, 53:73-84.
Oaxaca, R., 1973. ?Male?female Differentials in Urban Labor Markets,? International
Economic Review, 14: 693?709.
Peterson?s Guide to Four-Year Colleges (2002-2004) Peterson?s Guides. Princeton, NJ:
Peterson?s Guides.
Sigelman, Lee, and Samuel Bookheimer, 1983. ?Is It Whether You Win or Lose?
Monetary Contributions to Big-Time College Athletic Programs, ? Social Science
Quarterly, 64:347-359.
49
Sigelman, Lee, and Robert Carter, 1979. ?Win One for the Giver? Alumni Giving and
Big-Time College Sports, ? Social Science Quarterly, 60:284-294.
Spaeth, Joe. L., and Andrew M. Greeley, 1970. Recent Alumni and Higher Education.
New York: McGraw-Hill.
Tucker, Irvin B., 2004. ?A Reexamination of the Effect of Big-Time Football and
Basketball Success on Graduation Rates and Alumni Giving Rates,? Economics of
Education Review, 23:655-661.
U.S. Department of Education. Integrated Postsecondary Education Data System. Fiscal
years 2003-2004, nces.ed.gov/ipeds.
Wooldrige, Jeffrey, 2003. Introductory Econometrics: A Modern Approach. Mason, OH:
South-Western.
Yoo, J. H. and W. B. Harrison, 1989. ?Altruism in the Market for Giving and Receiving:
A Case of Higher Education,? Economics of Education Review, 8:367-376.
50
APPENDIX
TABLE I
Dependent Variable Information
Variable Description Mean Minimum Maximum
AlumTotal Total alumni contributions 7535183 2668 1.88 x10
8
AlumPS
Total alumni contributions
divided by total enrollment
1356.6 0.47 36979.83
AlumPA
Total alumni contributions
divided by one-quarter total
enrollment times upper bound
age
44.03 0.06 691.54
AlumPA2
Total alumni contributions
divided by one-quarter total
enrollment times age
10.57 0.02 230.99
51
TABLE II
Explanatory Variables
Variable Name Label
Expected
Direction
Description
Percent
Minority
%Min -
The ratio of minority (non-Caucasian)
students to total enrollment.
Percent Female %Fem -
The ratio of female students to total
enrollment.
Percent
Receiving
Student Loan
Aid
%RecLoan -
The percentage of student receiving
any type of student loan for fall 2003
to spring 2004.
Average
Amount Student
Loan Aid
AvgLoan -
The average amount of a student loan
at a respective university for the 2003-
2004 year.
Average SAT AvgSAT +
The average SAT score (or ACT
equivalent) of the fall 2003 incoming
freshmen class
Tuition Tuition +
Total tuition and fees for fall 2003.
Expenditure per
Student
StudExp
+
The ratio of total student expenditure
(instruction, student services, and
academic support) to total enrollment..
Enrollment Enroll +
Full-time enrollment for the 2003-204
academic year
Age of the
Institution
Age
+
Current year (2003) minus the
founding date of the school.
Private School
Dummy
Variable
Private +
A dummy variable indicating private
school status: ?O? if public institution,
?1? if private.
Tax Rate Tax
?
Average adjusted gross income tax rate
by state with applicable highest
marginal state tax
52
TABLE III
Explanatory Variable: Descriptive Statistics
45
Variable Mean Std Dev Minimum Maximum
%Min 0.29 0.18 0.03 0.99
%Fem 0.58 0.12 0.03 1
%RecLoan 0.57 0.19 0.04 1
AvgLoan 4133 1390 1002 11794
AvgSAT 1110 126.14 774 1520
Tuition 15558 7318 2554 30120
StudExp 12918 10323 3338 98628
Tax 0.19 0.03 0.15 0.32
Enroll 7501.24 8662.90 377 48397
Age 122.34 46.94 10 368
45
Private is not included because it is a dummy variable.
53
TABLE IV
Mean Values of Dependent Variables
Mean Values
Dependent Variable
Football Nonfootball
AlumTotal
1.09 x 10
7
2.68 x 10
6
AlumPS
1642.66
946.03
AlumPA
71.56
44.03
AlumPA2 47.19 35.27
54
TABLE V
Linear Regression Results with Differing Dependent Variables
Explanatory
Variable
AlumTotal
(I)
AlumPS
(II)
AlumPA
(III)
AlumPA2
(IV)
Intercept
-4.34 x 10
7
***
(-4.27)
-3263.47**
(-2.10)
-161.76**
(-2.37)
-28.78***
(-2.47)
Enroll
911.74***
(10.34)
Age
5.43 x 10
4
***
(4.05)
5.32**
(2.57)
AvgSAT
2.05 x 10
4
**
(2.62)
2.59**
(2.23)
0.14***
(2.82)
0.03***
(2.91)
Tuition
-360.92**
(-2.08)
0.04
(1.48)
1.57 x 10
-3
(1.34)
2.75 x 10
-4
(1.37)
%RecLoan
6.42 x 10
5
(0.02)
-1229.93**
(-2.09)
-46.81*
(-1.81)
-5.89
(-1.34)
AvgLoan
106.80
(0.25)
-0.23***
(-3.56)
-0.01***
(-3.48)
-1.56 x 10
-3
***
(-3.20)
%Min
7.01 x 10
6
*
(2.02)
-878.18*
(-1.69)
-41.24*
(-1.82)
-5.64
(-1.47)
%Fem
3.45 x 10
6
(0.72)
1894.18**
(2.56)
91.57***
(2.81)
19.37***
(3.49)
StudExp
724.27***
(8.89)
0.08***
(6.21)
3.35 x 10
-3
***
(6.12)
5.03 x 10
-4
***
(5.39)
Tax
9.64 x 10
5
(0.06)
-1729.51
(-0.70)
-48.24
(-0.44)
-12.02
(-0.65)
Private
8.05 x 10
6
**
(2.80)
533.66
(1.24)
27.75
(1.47)
4.38
(1.36)
Football
1.98 x 10
6
**
(1.55)
351.57*
(1.80)
19.61**
(2.35)
2.29*
(1.61)
Basketball
-4.94 x 10
5
(-0.19)
128.26
(0.32)
5.17
(0.29)
-0.48
(-0.16)
R
2
0.49 0.37 0.35 0.31
N 657 657 657 657
* Statistically significant at greater than 90% in a two-tailed test
** Statistically significant at greater than 95% in a two-tailed test
*** Statistically significant at greater than 99% in a two-tailed test
55
TABLE VI
Correlation Matrix
Explanatory
Variable
AvgSAT Tuition %RecLoan AvgLoan
AvgSAT 1 0.57 -0.27 0.19
Tuition 0.57 1 0.24 0.36
%RecLoan -0.27 0.24 1 0.30
AvgLoan 0.19 0.36 0.30 1
%Min -0.06 -0.03 -0.31 -0.13
%Fem -0.27 -0.01 0.13 -0.05
StudExp 0.72 0.59 -0.2 0.13
Tax 0.07 0.06 -0.16 -0.04
Private 0.29 0.87 0.46 0.38
Football 0.15 -0.03 0.01 -0.03
Basketball 0.08 0.03 0.07 0.08
56
TABLE VI (continued)
Correlation Matrix
Explanatory
Variable
%Min %Fem StudExp Tax
AvgSAT -0.06 -0.27 0.72 0.07
Tuition -0.03 -0.01 0.59 0.06
%RecLoan -0.31 0.13 -0.20 -0.16
AvgLoan -0.13 -0.05 0.13 -0.04
%Min 1 0.02 0.11 0.26
%Fem 0.02 1 -0.20 0.03
StudExp 0.11 -0.2 1 0.09
Tax 0.26 0.03 0.09 1
Private -0.13 0.11 0.36 -0.04
Football -0.21 -0.35 0.13 -0.15
Basketball -0.06 -0.08 0.02 -0.01
57
TABLE VI (continued)
Correlation Matrix
Explanatory
Variable
Private Football Basketball
AvgSAT 0.29 0.15 0.08
Tuition 0.87 -0.03 0.03
%RecLoan 0.46 0.01 0.07
AvgLoan 0.38 -0.03 0.08
%Min -0.13 -0.21 -0.06
%Fem 0.11 -0.35 -0.08
StudExp 0.36 0.13 0.02
Tax -0.04 -0.15 -0.01
Private 1 -0.11 -0.04
Football -0.11 1 0.23
Basketball -0.04 0.23 1
58
TABLE VII
Variance Inflation Factors
Explanatory
Variable
VIF
Intercept 0
AvgSAT 3.23
Tuition 8.23
%RecLoan 2.01
AvgLoan 1.26
%Min 1.36
%Fem 1.26
StudExp 2.53
Tax 1.12
Private 6.57
Football 1.33
Basketball 1.11
59
TABLE VIII
Heteroscedasticity Consistent Estimates
Explanatory
Variable
All Schools
Intercept
-161.76**
(-2.02)
AvgSAT
0.14***
(2.60)
Tuition
1.57 x 10
-3
(1.57)
%RecLoan
-46.81
(-1.65)
AvgLoan
-0.01***
(-3.86)
%Min
-41.24**
(-2.09)
%Fem
91.57**
(2.19)
StudExp
3.35 x 10
-3
***
(3.73)
Tax
-48.24
(-0.46)
Private
27.75**
(2.00)
Football
19.61**
(2.39)
Basketball
5.17
(0.38)
R
2
0.35
N 657
* Statistically significant at greater than 90% in a two-tailed test
** Statistically significant at greater than 95% in a two-tailed test
*** Statistically significant at greater than 99% in a two-tailed test
60
TABLE IX
Regression Results Seperated by Football
Explanatory
Variable
Total Sample
(I)
Football
(II)
Non-Football
(III)
Intercept
-128.33**
(-1.78)
-75.115
(-0.62)
-167.99**
(-1.88)
AvgSAT
0.15***
(2.71)
0.16*
(1.94)
0.14***
(2.66)
Tuition
1.74 x 10
-3*
(1.80)
1.2 x 10
-4
(0.10)
2.48 x 10
-3
(1.38)
%RecLoan
-38.73
(-1.43)
-39.75
(-1.21)
-48.28
(-1.22)
AvgLoan
-1.05 x 10
-2
***
(-4.29)
-9.51 x 10
-3
***
(-2.56)
-8.38 x 10
-3
***
(-3.34)
%Min
-51.97**
(-2.47)
-54.03*
(-1.79)
-18.95
(-0.76)
%Fem
68.51**
(1.79)
-90.45
(-1.08)
189.04***
(4.57)
StudExp
3.50 x 10
-3
***
(4.24)
2.90 x 10
-3
***
(2.93)
4.05 x 10
-3
***
(3.13)
Tax
-79.38
(-0.85)
91.13
(0.68)
-316.87**
(-2.27)
Private
20.56
(1.63)
58.45**
(2.57)
-13.49
(-0.92)
R
2
0.34 0.32 0.45
N 657 389 268
* Statistically significant at greater than 90% in a two-tailed test
** Statistically significant at greater than 95% in a two-tailed test
*** Statistically significant at greater than 99% in a two-tailed test
61
TABLE X
Decomposition
Percentage of Alumni Contributions
Differential Due to Football Programs
Explanatory
Variable
Alumni
Characteristics
Residual Component
AvgSAT
19.56
(2.66)
74.26
(0.17)
Tuition
-5.08
(1.38)
-121.23
(1.00)
%RecLoan
-0.77
(1.22)
17.86
(0.16)
AvgLoan
3.01
(3.34)
-16.78
(0.20)
%Min
5.46
(0.75)
-32.42
(0.77)
%Fem
-59.45
(4.57)
-549.20
(3.90)
StudExp
39.65
(3.13)
-58.86
(0.99)
Tax
12.45
(2.27)
274.03
(2.29)
Private
5.09
(0.92)
155.18
(1.94)
Total
(including intercept)
19.92 80.08
62
TABLE XI
Decomposition II
Percentage of Alumni Contributions
Differential Due to Non-Football Programs
Explanatory
Variable
Alumni
Characteristics
Residual Component
AvgSAT 22.11 71.71
Tuition -0.26 -126.06
%RecLoan -0.64 17.72
AvgLoan 3.42 -17.18
%Min 15.57 -42.53
%Fem 28.44 -637.09
StudExp 28.35 -47.56
Tax -3.58 290.06
Private -22.07 182.34
Total
(including intercept)
71.36 28.64
63
TABLE XII
Descriptive Statistics of Athletic Success Variables
46
Variable Mean Std Dev Minimum Maximum
WPFB
-2
0.53 0.25 0 1
WPFB
-1
0.51 0.24 0 1
WPFB 0.51 0.25 0 1
WPBB
-2
0.55 0.19 0.09 0.93
WPBB
-1
0.54 0.18 0.04 0.97
WPBB
0.53 0.19 0.04 0.94
46
FBPS
-2
, FBPS
-1
, FBPS, BBPS
-2
, BBPS
-1
and BBPS are not included because they are
dummy variables.
64
TABLE XIII
Mean Values of Variables by Division
Explanatory
Variable
All Divisions Division I Division II Division III
AlumPA 72.17 67.67 18.42 100.85
AvgSAT 1132.19 1152.41 1024.31 1158.4
Tuition 14204.04 10596.09 8062.73 20790.73
%RecLoan 0.57 0.45 0.64 0.66
AvgLoan 4136.55 3990.34 3764.73 4458.03
%Min 0.26 0.29 0.28 0.21
%Fem 0.54 0.53 0.57 0.54
StudExp 14355.06 14892.64 8028.16 16586.62
Tax 0.18 0.19 0.18 0.18
WPFB
-2
0.53 0.52 0.53 0.54
WPFB
-1
0.51 0.52 0.50 0.51
WPFB 0.51 0.51 0.51 0.50
WPBB
-2
0.54 0.54 0.55 0.54
WPBB
-1
0.54 0.54 0.55 0.54
WPBB
0.53 0.52 0.53 0.55
65
TABLE XIV
Regression Results with Differing Success Variables
Explanatory
Variable
All Divisions Division I Division II Division III
Intercept
-81.60
(-0.66)
-423.94
(-0.73)
-24.63
(-0.61)
-141.28
(-0.84)
AvgSAT
0.18**
(2.30)
0.29**
(2.21)
0.04
(1.38)
0.07
(0.57)
Tuition
-4.74 x 10
-4
(-0.34)
-4.39 x 10
-5
(-0.02)
-3.06 x 10
-3
**
(-2.53)
-1.74 x 10
-3
(-0.60)
%RecLoan
-51.95
(-1.27)
-1147.88*
(-1.78)
21.76
(1.35)
-36.87
(-0.52)
AvgLoan
-0.01***
(-2.77)
-0.01
(-1.08)
-0.01
(-1.51)
-0.01**
(-2.08)
%Min
-40.48
(-1.47)
-39.61
(-0.80)
-33.06**
(-2.14)
-80.17
(-1.25)
%Fem
-86.01
(-0.97)
264.92
(0.58)
-53.47*
(-1.84)
-162.24**
(-2.30)
StudExp
2.76 x 10
-3
***
(2.88)
1.93 x 10
-3
**
(2.14)
1.10 x 10
-3
(1.01)
4.50 x 10
-3
**
(2.24)
Tax
57.71
(0.42)
118.62
(0.59)
218.40**
(2.47)
-131.10
(-0.71)
Private
66.61**
(2.34)
77.35
(1.08)
60.36***
(3.82)
80.02
(1.56)
66
TABLE XIV (continued)
Regression Results with Differing Success Variables
Explanatory
Variable
All Divisions Division I Division II Division III
WPFB
-2
-2.97
(-0.14)
8.63
(0.25)
-18.27*
(-1.76)
15.98
(0.33)
WPFB
-1
30.09
(1.11)
80.64
(1.33)
8.05
(0.62)
-1.37
(-0.03)
WPFB
-18.47
(-0.79)
-80.30*
(-1.69)
-5.91
(-0.47)
-11.78
(-0.28)
FBPS
-2
29.87
(0.91)
68.63
(1.14)
10.80
(1.17)
-27.46
(-1.31)
FBPS
-1
-2.97
(-0.14)
-63.52*
(-1.84)
-1.98
(-0.17)
13.71
(0.71)
FBPS
30.09
(1.11)
-7.35
(-0.28)
-7.51
(-1.11)
2.34
(0.11)
WPBB
-2
-3.30
(-0.11)
-6.71
(-0.11)
-4.23
(-0.44)
-12.96
(-0.29)
WPBB
-1
-41.09
(-0.80)
-91.14
(-0.73)
38.79
(1.53)
-37.45
(-0.69)
WPBB
49.42
(1.05)
165.89
(1.17)
-28.21
(-1.55)
-1.00
(-0.03)
BBPS
-2
1.29
(0.11)
10.35
(0.56)
10.77*
(1.79)
-3.68
(-0.12)
BBPS
-1
-3.16
(-0.31)
13.07
(0.48)
-8.80
(-1.16)
-18.52
(-0.72)
BBPS
-14.95
(-0.85)
-70.56
(-1.53)
11.23
(1.63)
27.15
(0.73)
R
2
0.34 0.34 0.65 0.43
N 357 152 63 142
* Statistically significant at greater than 90% in a two-tailed test
** Statistically significant at greater than 95% in a two-tailed test
*** Statistically significant at greater than 99% in a two-tailed test
67
TABLE XV
F-Statistics on Division Regressions
All Divisions Division I Division II Division III
F-Statistic 0.72 0.93 0.78 0.39
? = 0.01 1.81 1.83 2.00 1.83
? = 0.05 2.30 2.34 2.65 2.34