Higher Education: School's Use of the Antitrust Exemption Has Not
Significantly Affected College Affordability or Likelihood of
Student Enrollment to Date (21-SEP-06, GAO-06-963).
In 1991 the U.S. Department of Justice sued nine colleges and
universities, alleging that they had restrained competition by
making collective financial aid determinations for students
accepted to more than one of these schools. Against the backdrop
of this litigation, Congress enacted a temporary exemption from
antitrust laws for higher education institutions in 1992. The
exemption allows limited collaboration regarding financial aid
practices with the goal of promoting equal access to education.
The exemption applies only to institutional financial aid and can
only be used by schools that admit students without regard to
ability to pay. In passing an extension to the exemption in 2001,
Congress directed GAO to study the effects of the exemption. GAO
examined (1) how many schools used the exemption and what joint
practices they implemented, (2) trends in costs and institutional
grant aid at schools using the exemption, (3) how expected family
contributions at schools using the exemption compare to those at
similar schools not using the exemption, and (4) the effects of
the exemption on affordability and enrollment. GAO surveyed
schools, analyzed school and student-level data, and developed
econometric models. GAO used extensive peer review to obtain
comments from outside experts and made changes as appropriate.
-------------------------Indexing Terms-------------------------
REPORTNUM: GAO-06-963
ACCNO: A61250
TITLE: Higher Education: School's Use of the Antitrust Exemption
Has Not Significantly Affected College Affordability or
Likelihood of Student Enrollment to Date
DATE: 09/21/2006
SUBJECT: Antitrust law
College students
Colleges and universities
Comparative analysis
Cost analysis
Financial analysis
Higher education
Income statistics
Price fixing
Program evaluation
Student financial aid
******************************************************************
** This file contains an ASCII representation of the text of a **
** GAO Product. **
** **
** No attempt has been made to display graphic images, although **
** figure captions are reproduced. Tables are included, but **
** may not resemble those in the printed version. **
** **
** Please see the PDF (Portable Document Format) file, when **
** available, for a complete electronic file of the printed **
** document's contents. **
** **
******************************************************************
GAO-06-963
* Results in Brief
* Background
* Legal History of Antitrust Exemption for Higher Education In
* Determining a Student's Financial Need
* Twenty-Eight Schools Used the Antitrust Exemption to Develop
* Highly Selective Private4-Year Colleges and Universities For
* Participating Schools Agreed to a Common Methodology for Ass
* As the Cost of Attendance at Schools Using the Exemption Ros
* Cost of Attendance Increased at Schools Using the Exemption
* Percentage of Students Receiving Institutional Grant Aid and
* Students Accepted to Both Schools Using the Exemption and Co
* Implementation of a Common Methodology Has Not Significantly
* Concluding Observations
* Agency Comments
* Appendix I: Statistical Analysis of Expected Family Contribu
* Appendix II: Econometric Analysis of Effects of the Higher E
* Theories of the Effects of the Consensus Approach on Financi
* Sources of Data for the Model
* Selection of Control Schools
* Determination of the Appropriate Time Periods for Assessing
* Specifications of Econometric Models and Estimation Methodol
* Modeling the Effects of the Consensus Approach Methodology f
* Comparison of Prices and Financial Aid in CA and Non-CA Scho
* Model Specifications and Estimation Methodology
* Estimation Results of the Effects of Attending Meetings and
* Total Effects of Implementing the Consensus Approach (from t
* Prior Levels and Differential Effects of the Consensus Appro
* Limitations of the Study
* Sample Selection bias
* Measures of Price
* Early Decision Admissions
* Excluded Schools of Comparable Selectivity
* Limited Data Availability
* Appendix III: Classification of 1999-2000 Academic Year and
* Does Academic Year 1999-2000 belong to the Pre- or Post- Con
* Do the Schools That Only Attended the 568 group Meetings be
* To Which Group Did Brown Belong-Control or Treatment?
* To Which Group Did Stanford Belong-Control or Treatment?
* To Which Group Did USC Belong-Control or Treatment?
* Appendix IV: Comments from 568 Presidents' Group
* Appendix V: Consultants and Peer Reviewers
* Appendix VI: GAO Contact and Staff Acknowledgments
* Bibliography
* Order by Mail or Phone
Report to Congressional Committees
United States Government Accountability Office
GAO
September 2006
HIGHER EDUCATION
Schools' Use of the Antitrust Exemption Has Not Significantly Affected
College Affordability or Likelihood of Student Enrollment to Date
GAO-06-963
Contents
Letter 1
Results in Brief 4
Background 6
Twenty-Eight Schools Used the Antitrust Exemption to Develop a Common
Methodology for Assessing a Family's Financial Need 9
As the Cost of Attendance at Schools Using the Exemption Rose, the Amount
of Institutional Grant Aid They Provided to Students Increased at a Slower
Rate 15
Students Accepted to Both Schools Using the Exemption and Comparable
Schools Had No Appreciable Difference in the Amount They Would Be Expected
to Contribute Towards College 19
Implementation of a Common Methodology Has Not Significantly Affected
Affordability or Enrollment at Schools Using the Exemption 20
Concluding Observations 23
Agency Comments 23
Appendix I Statistical Analysis of Expected Family Contributions at
Schools Using the Exemption and Comparable Schools 27
Appendix II Econometric Analysis of Effects of the Higher Education
Antitrust Exemption on College Affordability and Enrollment 32
Theories of the Effects of the Consensus Approach on Financial Aid 32
Sources of Data for the Model 33
Selection of Control Schools 34
Specifications of Econometric Models and Estimation Methodology 38
Estimation Results of the Effects of Attending Meetings and Implementing
the Consensus Approach 62
Limitations of the Study 70
Appendix III Classification of 1999-2000 Academic Year and Schools Only
Attending the 568 Group Meetings 74
Does Academic Year 1999-2000 belong to the Pre- or Post- Consensus
Approach Implementation Period? 74
Do the Schools That Only Attended the 568 group Meetings belong to the
Control or Treatment Group? 75
Appendix IV Comments from 568 Presidents' Group 77
Appendix V Consultants and Peer Reviewers 87
Appendix VI GAO Contact and Staff Acknowledgments 88
Bibliography 89
Tables
Table 1: Comparison of the Federal Methodology and the College Board's
Base Institutional Methodology for Need Analysis 9
Table 2: Schools Using the Antitrust Exemption, as of May 2006 10
Table 3: Comparison of Consensus Approach Developed by Schools Using the
Antitrust Exemption Compared to the College Board's Institutional
Methodology 13
Table 4: Number of Schools That Did Not Implement Certain Consensus
Approach Options in School Year 2005-2006 14
Table 5: Estimated Changes in Amount Paid, Financial Aid, and Enrollment
at Schools Using the Consensus Approach Compared to Schools Not Using the
Exemption 21
Table 6: Schools Included in Analysis of Effects of Exemption 23
Table 7: Summary Statistics of Expected Family Contributions 29
Table 8: Tests of Variations in Expected Family Contributions 31
Table 9: Control and Treatment Schools for Analyzing Effects of the
Consensus Approach Implementation 37
Table 10: Summary Statistics of Variables Used in Regression Analysis,
1995-1996 and 2003-2004: CA Schools 45
Table 11: Summary Statistics of Variables Used in Regression Analysis,
1995-1996 and 2003-2004: Non-CA Schools 46
Table 12: CA and Non-CA Schools: Price and Financial Aid 48
Table 13: CA and Non-CA Schools-Financial Aid Applicants Only: Price and
Financial Aid 49
Table 14: Regression Estimates of Effects of Consensus Approach
Implementation on Price, Tuition, and Enrollment 59
Table 15: Regression Estimates of Effects of Consensus Approach
Implementation on Financial Aid 61
Table 16: Estimates of Effects of Consensus Approach Implementation on
Affordability and Enrollment in CA Schools Relative to Non-CA Schools 65
Table 17: Estimates of Affordability and Enrollment before the Consensus
Approach Implementation for Particular Groups of Students in Both CA and
Non-CA Schools 68
Table 18: Differential Effects of Consensus Approach Implementation on
Affordability and Enrollment in CA Schools for Particular Groups of
Students 69
Table 19: Comparison of Observed and Predicted Price and Financial Aid
Variables in CA and Non-CA Schools: Pre- and Post-Consensus Approach
Implementation Period 72
Figures
Figure 1: Determining a Student's Financial Need 8
Figure 2: Average Tuition, Fees, and Room and Board at Schools Using the
Antitrust Exemption Compared to All Other Private 4-Year Not-For-Profit
Schools and Comparable Schools, School Years 2000 to 2005 16
Figure 3: Percentage of Students at Schools Using the Antitrust Exemption
Receiving Various Types of Institutional Grant Aid from 2000 to 2006 17
Figure 4: Average Amount of Various Institutional Grant Aid Awards at
Schools Using the Antitrust Exemption from 2000 to 2006 18
Abbreviations
CA Consensus Approach
EFC expected family contribution
IPEDS Integrated Postsecondary Education Data System
MIT Massachusetts Institute of Technology
NPSAS National Postsecondary Student Aid Study
This is a work of the U.S. government and is not subject to copyright
protection in the United States. It may be reproduced and distributed in
its entirety without further permission from GAO. However, because this
work may contain copyrighted images or other material, permission from the
copyright holder may be necessary if you wish to reproduce this material
separately.
United States Government Accountability Office
Washington, DC 20548
September 21, 2006
Congressional Committees
In 1991, the U.S. Department of Justice sued nine colleges and
universities, alleging that by collectively making financial aid
determinations for students accepted to more than one of these schools,
the schools had unlawfully conspired to restrain trade in violation of the
Sherman Act. Specifically, Justice argued that by agreeing upon the amount
of money that the families of admitted students would be expected to pay
towards their student's education, these schools were engaging in price
fixing. Justice and the schools ultimately reached settlements that ended
these activities. These schools, which are among the nation's most
prestigious private universities, had engaged in these activities for more
than 30 years.
Against the backdrop of this litigation, in 1992 Congress enacted a
temporary exemption from the antitrust laws for higher education
institutions that has been renewed several times and is set to expire in
2008. Under the exemption, schools are allowed a limited degree of
collaboration on financial aid practices in the hope that it would further
the government's goal of promoting equal access to educational
opportunities for students, including low income and minority students.
Under the exemption, schools that admit students without regard to ability
to pay would be able to develop and use common principles of financial aid
policies and make changes to formulas used to calculate financial aid
awards, but not discuss specific students' awards. Specifically, such
schools would be allowed to engage in the following joint practices:
1. agreeing to award financial aid only on the basis of
demonstrated financial need;
2. using common principles of analysis for determining financial
need;
3. using a common aid application form; and
4. exchanging, through an independent third party, financial
information submitted by students and their families.
The exemption only applies to an institution's own aid. Federal aid, which
is allocated based on a statutory formula, was not targeted by the
exemption. Proponents of the exemption believe that common principles
could lead to a more equitable allocation of aid, make attendance at
schools using the exemption more affordable, and, in turn, increase
enrollment of low income students at these schools. Moreover, proponents
believe that allowing schools to use common principles for determining
financial need should reduce variation among schools in what a family is
expected to pay and enable students to choose a school without making cost
the defining factor. On the other hand, some are concerned that exempting
schools from antitrust laws would reduce competition. Specifically, with
less competition, some students would pay more for college because their
opportunities to consider price differences when choosing schools would be
diminished.
In passing the 2001 extension to the exemption, Congress directed GAO to
study whether the exemption resulted in changes in the amount students and
their families would pay for college. In response to this mandate, we
determined: (1) how many schools used the exemption and what joint
practices these schools implemented, (2) trends in cost of attendance and
institutional grant aid at schools using the exemption, (3) how expected
family contributions at schools using the exemption compare to those at
similar schools that did not use the exemption, and (4) the effects of the
exemption on affordability and enrollment.
To determine the number of schools that made use of the exemption since
1992, we reviewed literature and studies on the exemption, interviewed
higher education associations, and reviewed documents that identified a
group of schools that were using the exemption. We interviewed officials
of these schools, reviewed reports of their activities, and collected
information on their financial aid policies. To determine if other schools
might have formed groups to participate in activities allowed under the
exemption, we also surveyed selected similar schools and found no such
other groups.
To determine trends in cost of attendance-tuition, room, and board-and
institutional grant aid at the schools using the exemption, we collected
data from them and supplemented it with information available from the
U.S. Department of Education's Integrated Postsecondary Education Data
System (IPEDS) for school years 2000-2001 through 2005-2006. We received
data from 26 of the 28 schools using the exemption. We determined that the
IPEDS and institutional data were sufficiently reliable and valid for
purposes of our review.
To determine how expected family contributions (EFC) at schools using the
exemption compared to those at similar schools not using the exemption, we
collected and compared student-level EFC data from both sets of schools as
of April 1, 2006. To assess the extent of variation in EFC across multiple
schools, we isolated the EFCs of individual students accepted at (1)
multiple schools using the exemption, (2) multiple schools not using the
exemption, and (3) both schools using the exemption and schools that did
not. While EFC determinations of students accepted at both schools using
the exemption and those that did not best show the extent of variation
because it allows us to control for differences in student
characteristics, this group of students was small. Thus, we supplemented
our analysis with data from the other two groups listed. Based on our
discussions with school officials on the steps taken to ensure reliability
of the EFC data, we determined that the data were sufficiently reliable
and valid for purposes of our review. See appendix I for further details
of our statistical analysis.
To assess the effects of the exemption on affordability and enrollment, we
developed econometric models to examine the effects of the exemption on
tuition, financial aid (including grants and loans), amount paid for
college (measured by the total cost of attendance less total grant aid),
and student enrollment at schools using the exemption. Determining
"effect" requires both a treatment group (those schools using the
exemption) and a control group (a comparable set of schools that did not
use the exemption) as well as controlling for variations in the actions of
the schools over time that are independent of the exemption. Differences
found between the two groups in terms of affordability and enrollment
(effects) can then be attributed to the exemption (treatment). GAO's
econometric analysis was focused on the mandate from Congress that
requires us to examine the effects of the exemption. It is different from
a market-specific analysis conducted in an antitrust investigation and is
not intended to address whether or not conduct may be taking place that
might violate the antitrust laws in the absence of the exemption. In order
to find a comparative set of schools, we used the U.S. News and World
Report annual rankings of the "best colleges." We obtained school-level
data from IPEDS and student-level data from the National Postsecondary
Student Aid Study for academic years 1995-1996, 1999-2000, and 2003-2004.
We also collected data from other sources, including a GAO survey of the
schools using the exemption and the comparable schools. We analyzed
whether there were any effects on affordability and enrollment at schools
using the exemption for all students and whether there were differences by
family income or race. We also controlled for other factors that could
cause changes in affordability and enrollment, such as school or student
characteristics. Because of data limitations, we were not able to include
all schools using the exemption in the treatment group. Nevertheless,
there were sufficient similarities between the excluded schools and the
schools we included in our model to allow for a meaningful analysis. In
developing the models, we reviewed several studies on the economics of
higher education. We provided a detailed draft outline of our econometric
methodology, including a description of the types and sources of data we
used, to outside experts with whom we consulted on the design and analysis
because of their in-depth knowledge of antitrust law and the economics of
higher education. We also provided a draft of our report to peer reviewers
in academia and incorporated their comments when appropriate. See
appendixes II and III for a detailed explanation of our econometric
analysis. We conducted our work in accordance with generally accepted
government auditing standards between May 2005 and September 2006.
Results in Brief
Twenty-eight schools-all highly selective, private 4-year
institutions-formed a group to use the antitrust exemption, and of the
four collaborative activities allowed, the group has engaged in only
one-development of a common methodology for assessing financial need,
which the group called the "consensus approach." With respect to the other
three activities allowed under the exemption, the schools either chose not
to engage in the activities or piloted them on a limited basis. For
example, three schools in the group attempted to share student-level
financial aid data through a third party. However, the schools reported
that because the effort was too burdensome and yielded little useful
information, they chose not to continue. The consensus approach to need
analysis developed by the group is based on elements already a part of
another need analysis methodology that considers a family's income and
assets to determine a student's ability to pay for college. Schools
modified some elements of that methodology and reached agreement on how to
define those elements. Although schools in the group agreed to the concept
of the consensus approach, the schools varied in their implementation of
the methodology. Schools that partially implemented or did not implement
the consensus approach often cited concerns about potential increased
costs associated with implementing the methodology. Twenty five of the 28
schools implemented the consensus approach methodology; three did not.
Schools that chose to use part or all of the elements of the consensus
approach did so between 2002 and 2005.
Over the last 5 years, tuition, room, and board costs at the group of
schools using the exemption increased, and while the amount of grant aid
these schools provided to students also increased, it did so at a slower
rate. Between school years 2000-2001 and 2004-2005, tuition, room, and
board increased by 13 percent, from $38,319 to $43,164, compared to a 7
percent increase at other private 4-year not-for-profit schools. Average
institutional grant aid awards increased by 7 percent from $18,675 to
$19,901 at schools using the exemption, and the percentage of students
receiving such aid increased from 37 to 40 percent, from school years
2000-2001 to 2005-2006. Among students receiving institutional grant aid
awards, the percent of students who received need-based institutional
grant aid at schools using the exemption increased from 34 to 36 percent,
and the percent of students receiving non-need-based institutional grant
aid awards (i.e., academic or athletic scholarships) also increased
slightly from 2 to 4 percent.
We found virtually no difference in the amounts students and their
families were expected to pay at schools using the exemption compared to
similar schools not using the exemption. Average expected family
contribution (EFC) for students accepted at schools using the exemption
was $27,166 and for those accepted at comparable schools not using the
exemption was $27,395 in school year 2005-2006. While officials from
schools using the exemption expected that students accepted to several of
their schools would experience less variation in their EFC, we found that
the variation in the EFC for a student who was accepted to several schools
using the exemption was similar to the variation in EFC that same student
received from schools not using the exemption. The variation in EFCs for
these students was about $6,000 at both sets of schools. Not all schools
using the consensus approach chose to adopt all the elements of the
methodology, a factor that may account for the lack of consistency in EFCs
among schools using the exemption. For example, seven schools chose not to
use the consensus method for considering home equity that could have
contributed to the variation in EFCs at schools using the exemption.
Based on our analysis, schools' use of the consensus approach did not have
a significant impact on affordability-the amount students and families
paid for college, which is measured by the total cost of attendance less
total grant aid-or affect the likelihood of enrollment at schools using
the exemption. While we found that the use of the consensus approach
resulted in higher amounts of need-based grant aid awarded to some student
groups (middle income, Asian students, and Hispanic students) compared to
their counterparts at schools not using the consensus approach, the total
amount of grant aid awarded did not significantly change. It is likely
that because the change in total grant aid was similar compared to the
change at schools not using the consensus approach, the increase in
need-based grant aid was offset by a decrease in non-need-based aid, such
as academic scholarships. We also found that low income students at
schools using the consensus approach, compared to those at schools not
using the consensus approach, received a significantly higher amount of
total aid, which includes both grants and loans. However, the amount of
grant aid that these students received did not significantly change, which
suggest they likely received more aid in the form of loans, which they
would need to repay. Additionally, implementing the consensus approach did
not affect the likelihood of low-income or minority students enrolling at
schools using the consensus approach compared to schools that did not.
Because we have data for only one year after implementation, it is
possible that some eventual effects of the consensus approach may not be
captured. The effects of using the consensus approach could be gradual,
rather than immediate, and therefore may not be captured until later
years.
We provided the group of schools using the antitrust exemption, Secretary
of Education, and Attorney General with a copy of our draft report for
review and comments. The group of schools using the exemption reviewed a
draft of this report and stated it was a careful and objective report, but
raised concerns about the data used in our econometric analysis and the
report's tone and premise. We believe that the data we used were reliable
to support our conclusions. The group of schools using the exemption also
provided technical comments, which we incorporated where appropriate. The
group's written comments appear in appendix IV. The Department of
Education reviewed the report and did not have any comments. The
Department of Justice provided technical comments, which we incorporated
where appropriate.
Background
Legal History of Antitrust Exemption for Higher Education Institutions
In the early 1990's the U.S. Department of Justice (Justice) sued nine
universities and colleges, alleging that their practice of collectively
making financial aid decisions for students accepted to more than one of
their schools restrained trade in violation of the Sherman Act.1 By
consulting about aid policies and aid decisions, through what was known as
the Overlap group, the schools made certain that students who were
accepted to more than one Overlap school would be expected to contribute
the same towards their education. Thus, according to Justice, "fixing the
prices" students would be expected to pay. All but one school,
Massachusetts Institute of Technology (MIT), settled with Justice out of
court, ending the activities of the Overlap group. The District Court
ruled that MIT's joint student aid decisions in the Overlap group violated
the Sherman Act. On appeal, the Third Circuit Court of Appeals agreed with
the District Court that the challenged practices were commercial activity
subject to the antitrust laws. However, it reversed the judgment and
directed the District Court to more fully consider the procompetitive and
noneconomic justifications advanced by MIT during the court proceedings
and whether social benefits attributable to the practices could have been
achieved by means less restrictive of competition.2 In recognition of the
importance of financial aid in achieving the government's goal of
educational access, but also mindful of the importance of antitrust laws
in ensuring the benefits of competition, the Congress passed a temporary
antitrust exemption.3 In 1994, Congress extended the exemption and
specified the four collective activities in which schools that admit
students on a need-blind basis could engage.4 The exemption was extended
most recently in 2001, and is set to expire in 2008.5
1The schools sued were: Brown University, Columbia University, Cornell
University, Dartmouth College, Harvard College, Massachusetts Institute of
Technology, Princeton University, University of Pennsylvania, and Yale
University.
Determining a Student's Financial Need
For many students, financial aid is necessary in order to enroll in and
complete a postsecondary education. In school year 2004-2005, about $113
billion in grant, loan, and work-study aid was awarded to students from a
variety of federal, state, and institutional sources.6 Need analysis
methodologies are used to determine the amount of money a family is
expected to contribute toward the cost of college and schools use this
information in determining how much need-based financial aid they will
award. For the purposes of awarding federal aid, expected family
contribution (EFC) is defined in the Higher Education Act of 1965, as
amended, as the household financial resources reported on the Free
Application for Federal Student Aid, minus certain expenses and
allowances. The student's EFC is then compared to the cost of attendance
to determine if the student has financial need. (see fig. 1)
2U.S. v. Brown, 5 F.3d 658 (3rd Cir. 1993). The Department of Justice and
MIT subsequently entered into a settlement agreement in which MIT agreed
to certain "Standards of Conduct."
3Pub. L. No. 103-325 (1992).
4Pub. L. No. 103-382 (1994).
5Pub. L. No. 107-72 (2001).
6Some financial aid is awarded to students based on merit rather than
financial need.
Figure 1: Determining a Student's Financial Need
While the federal methodology is used to determine a student's eligibility
for federal aid, some institutions use this methodology to award their own
institutional aid. Others prefer a methodology developed by the College
Board (called the institutional methodology) or their own methodology.7
Schools that use the institutional methodology require students to
complete the College Scholarship Service/Financial Aid PROFILE application
and the College Board calculates how much they and their families will be
expected to contribute toward their education. Schools that use these
alternative methodologies feel they better reflect a family's ability to
pay for college because they consider many more factors of each family's
financial situation than the federal methodology. For example, the
institutional methodology includes home and farm equity when calculating a
family's ability to pay for college, while the federal methodology
excludes them. See table 1 below for a comparison of the federal
methodology to the institutional methodology.
7The College Board is a not-for-profit membership association composed of
more than 5,000 schools, colleges, universities, and other educational
organizations. In conjunction with financial aid professionals and
economists, the College Board developed its own methodology to measure a
family's ability to pay for college.
Table 1: Comparison of the Federal Methodology and the College Board's
Base Institutional Methodology for Need Analysis
Federal methodology Institutional methodology
Home equity Not included. Included.
Family farm equity Not included. Included.
Student assets Included, 35 percent of Included, 25 percent of
student's net worth student's net worth
expected to be used for expected to be used for
college costs. Minimum college costs. Minimum
contribution from contribution from student
student expected. expected.
Family assets Excluded the assets of Included a fuller range of
families whose income family assets, such as
fell below $50,000 and home equity, other real
who filed a simple tax estate, and business and
return. farm assets.
12 percent of assets 5 percent of assets
expected to be used expected to be used
towards college. towards college.
Divorced and separated Excluded noncustodial Included noncustodial
families parent income and parent income and assets.
assets.
(Noncustodial parent
contribution)
Total income Included only the Included in total income
adjusted gross income any untaxed income and any
reported on federal tax paper depreciation and
returns, plus various business, rental, or
categories of untaxed capital losses that
income. artificially reduced
adjusted gross income.
Medical/elementary and Not included. Included.a
secondary school
expenses
Cost of living variance Not included. Not included.b
Number of siblings in Included-divides the Included-instead of
college parental contribution dividing by the number in
by the number of college, parental
siblings enrolled in contribution per student
college. reduced by 40 percent for
2 in college and by 55
percent for 3.
Source: GAO analysis.
Notes: The institutional methodology is the base one provided by the
College Board to schools. A school may select other options available in
the institutional methodology when assessing a student's financial need.
aElementary and secondary school expenses are an option that could be
added by a school.
bCost of living variance is an option that could be used by a school.
Twenty-Eight Schools Used the Antitrust Exemption to Develop a Common
Methodology for Assessing a Family's Financial Need
Twenty-eight schools formed a group under the antitrust exemption and
engaged in one of the four activities allowable under the exemption.
School officials believed that the one activity-development of a common
methodology for assessing financial need-would help reduce variation in
amounts students were expected to pay when accepted to multiple schools
and allow students to base their decision on which school to attend on
factors other than cost. In developing the common methodology, called the
consensus approach, schools modified an existing need analysis methodology
and reached agreement on how to treat each element of the methodology.
While the schools reached agreement on a methodology, implementation of
the methodology among the schools varied.
Highly Selective Private4-Year Colleges and Universities Formed a Group to
Participate in Activities Allowable under the Exemption
Twenty-eight schools, all of which have need-blind admission policies as
required under the law, formed the 568 Presidents' Group in 1998 with the
intent to engage in activities allowed by the antitrust exemption.8
Members of the group are all private 4-year schools that have highly
selective admissions policies. One member school dropped out of the group
because the school no longer admitted students on a need-blind basis. (See
table 2 below for a list of current and former member schools.)
Table 2: Schools Using the Antitrust Exemption, as of May 2006
Amherst College Middlebury College
Boston College Northwestern University
Brown University Pomona College
Claremont McKenna College Rice University
Columbia University Swarthmore College
Cornell University University of Chicago
Dartmouth College University of Notre Dame
Davidson College University of Pennsylvania
Duke University Vanderbilt University
Emory University Wake Forest University
Georgetown University Wellesley College
Grinnell College Wesleyan University
Haverford College Williams College
Massachusetts Institute of Technology Yale University
Source: GAO analysis.
Note: Bowdoin College and Macalester College were once members of the
group.
Membership is open to colleges and universities that have need blind
admissions policies in accordance with the law. Member schools must (1)
sign a certificate of compliance confirming the institution's need-blind
admissions policy and (2) submit a signed memorandum of understanding that
indicates willingness to participate in the group and adhere to its
guidelines. Additionally, members share in paying the group's expenses.
In addition to the group's 28 members, 6 schools attended meetings of the
group to observe and listen to discussions, but have not become members.9
In order to attend meetings, observer schools were required to provide a
certificate of compliance stating that they had a need-blind admission
policy. Observer schools explained that their participation was based on a
desire to be aware of what similar schools were thinking in terms of need
analysis methodology, as well as have an opportunity to participate in
these discussions. Despite these benefits, observer schools said they
preferred not to join as members because they did not wish to agree to a
common approach to need analysis or they did not want to lose
institutional independence.
8568 refers to the section in the Improving America's Schools Act of 1994
where the exemption is contained.
Other institutions with need-blind admissions reported that, although
eligible to participate in activities allowed by the exemption, they were
not interested or not aware of the group formed to use the antitrust
exemption. Some told us that they did not understand how students would
benefit from the schools' participation in such activities. Others cited
limited funding to make changes to their need analysis methodology and
concerns that they would lose the ability to award merit aid to
students.10
Participating Schools Agreed to a Common Methodology for Assessing Financial
Need, but Schools Varied in Their Implementation of the Methodology
Of the four activities allowed under the antitrust exemption, the 28
schools engaged in only one-development of the consensus approach for need
analysis. With respect to the other three activities allowed under the
exemption, the schools either chose to not engage in the activities or
piloted them on a limited basis. For example, three schools in the group
attempted to share student-level financial aid data through a third party.
However, they reported that because the effort was too burdensome and
yielded little useful information, they chose not to continue. The group
also expressed little need or interest in creating another common aid
application form as such a form already existed. Schools also decided to
leave open the option to award aid on a non-need basis.
According to the officials representing the 28 schools, the main purpose
of the group was to discuss ways to make the financial aid system more
understandable to students and their families and commit to developing a
common methodology for assessing a family's ability to pay for college,
which they called the consensus approach. Developing an agreed upon common
approach to need analysis, according to school officials, might help
decrease variation in what families were expected to pay when accepted to
multiple schools, allowing students to base their decision on what school
to attend on factors other than cost. School officials also believed that
agreeing to a common need analysis methodology would produce expected
family contributions that were reasonable and fair for families and allow
schools to better target need-based aid. The group did not address the
composition of a student's financial aid package; specifically, what
combination of grants, loans, or work-study a student would receive.
9These schools were: California Institute of Technology, Case Western
University, Harvard University, Stanford University, Syracuse University,
and University of Southern California.
10Participation in the 568 Presidents' Group, however, does not prohibit
members from awarding merit aid.
In developing the consensus approach for need analysis, the schools
modified elements already in the College Board's institutional
methodology, but member schools agreed to treat these elements the same
when calculating a student's EFC. Some of the modifications that the group
made to College Board's institutional methodology were later incorporated
into the institutional methodology. The consensus approach and the
institutional methodology similarly treat income from the non-custodial
parent, and both account for the number of siblings in college in the same
manner when calculating a student's expected family contribution. However,
there are differences in how each methodology treats a family's home
equity and a student's assets. For example, the institutional methodology
uses a family's entire home equity in its assessment of assets available
to pay for college, while the consensus approach limits the amount of home
equity that can be included. According to one financial aid officer at a
member school, including the full amount of a family's home equity was
unfair to many parents because in some areas of the country the real
estate market had risen so rapidly that equity gains inflated a family's
assets. Officials representing some member schools stated that adjustments
to home equity would likely affect middle and upper income families more
than lower income families who are less likely to own a home. Table 3
below further illustrates the differences and similarities between the
consensus approach and the institutional methodology.
Table 3: Comparison of Consensus Approach Developed by Schools Using the
Antitrust Exemption Compared to the College Board's Institutional
Methodology
Institutional methodology Consensus approach
Home equity Included. No limit on Included. Home value
amount considered asset is capped at 2.4 times
available to pay for income minus mortgage
college. debt.
Family farm equity Included. Included.
Student and family Included, but assets Included. In general
assets counted separately. student assets-such as
prepaid and college
25 percent of student's net savings plans are
worth expected to be used combined with family
for college costs. assets. 5 percent of
family assets expected
5 percent of parent's to be used for
assets expected to be used college. Trust funds
for college costs. will be considered on
a case by case basis.
Divorced and separated Included. Expects Same as IM.
families noncustodial parent to
contribute towards college
(Noncustodial parent) costs.
Total income/adjusted Included in total income Excluded business and
gross income any untaxed income and any rental losses from
paper depreciation and calculation of income.
business, rental, or
capital losses which
artificially reduced
adjusted gross income.
Medical/elementary and Included.a Included.
secondary school
expenses
Cost of living variance Excluded.b Adjusted living
expenses based on
geographic location.
Takes into
consideration that it
is more costly to live
in some areas of the
country.
Number of siblings in Included-considers number Same as IM.
college of children enrolled in
college, but instead of
dividing by the number in
college, it reduced the
parental contribution for
each student by 40 percent
if 2 in college and by 55
percent if 3.
One-time income Not included.c Excluded income that
adjustment was not received on an
annual basis, such as
unemployment income or
capital gains.
Family debt Not included. Made allowance for
debt payments on loans
incurred by parents
for student's
education.
Source: GAO analysis.
Note: The consensus approach is being compared to the base institutional
methodology. Schools may choose to implement other options available under
the institutional methodology when assessing a student's financial need.
aPrivate elementary and secondary school tuition allowed at the option of
the institution.
bAs an option schools can adjust living expenses based on geographic
locations.
cThis is not in the base IM; however, a financial aid officer can adjust
for this on a case-by-case basis, consistent with professional judgment.
In addition, under the consensus approach schools agreed to a common
calendar for collecting data from families. Members continue to maintain
the ability to exercise professional judgment in assessing a family's
ability to pay when there are unique or extenuating financial
circumstances.
Twenty-five of 28 schools implemented the consensus approach; 3 did not.
While 13 schools implemented all the elements of the consensus approach,
the remaining schools varied in how they implemented the methodology. As
shown in table 4 below, seven schools chose not to use the consensus
approach method for accounting for family loan debt, home equity, and
family and student assets.
Table 4: Number of Schools That Did Not Implement Certain Consensus
Approach Options in School Year 2005-2006
Number of schools that did not
Options in the consensus approach implement optiona
Number of siblings in college 1
One-time income adjustments 2
Elementary and secondary school tuition
expenses 3
Medical expenses 3
Cost of living variances 5
Divorced and separated families 6
Family and student assets 7
Home equity 7
Family loan debt 7
Source: GAO analysis of schools' survey responses.
aA total of 25 member schools used part or all of the consensus approach.
The 25 schools that implemented the consensus approach did so between 2002
and 2005. Member schools reported that they preferred to use the consensus
approach as opposed to other available need analysis methodologies because
it was more consistent and fairer than alternative methodologies.
Moreover, according to institution officials, they believed the new
methodology had not reduced price competition and had resulted in the
average student receiving more financial aid. In some cases, if using the
consensus approach lowered a student's EFC, the institution would then
allocate more money for financial aid than it would have if it had used a
different need analysis methodology. For some schools the consensus
approach was not that different from the methodology their institution
already had in place, but other schools said that fully implementing the
consensus approach cost their school more money. Among schools that
partially implemented the consensus approach, many explained they did not
fully implement the new methodology because it would have been too costly.
As the Cost of Attendance at Schools Using the Exemption Rose, the Amount of
Institutional Grant Aid They Provided to Students Increased at a Slower Rate
The cost to attend the schools participating under the exemption rose over
the past 5 years by over 10 percent while cost increases at all other
private schools rose at about half that rate. At the same time, the
percentage of students receiving institutional aid increased and
institutions increased the amount of such aid they provided students,
although at a slower rate than cost increases.
Cost of Attendance Increased at Schools Using the Exemption Corresponding to
Increases at Other Private Schools
During the past 5 years, the cost of attendance-tuition, fees, room, and
board-at schools using the exemption increased by approximately 13 percent
from $38,319 in school year 2000-2001 to $43,164 in school year 2004-2005,
a faster rate than other schools.11 For example, at other private 4-year
schools there was a 7 percent increase in these costs, from $25, 204 to
$27, 071.12 Additionally, as figure 2 illustrates, among a set of schools
that were comparable to the schools using the exemption, costs increased
by 9 percent from $40,238 to $43,939 over that same time period.13
11All dollar amounts are in 2005 dollars. Data presented for schools using
the exemption was collected from 26 of the 28 schools using the exemption.
12Other private 4-year schools include not-for-profit institutions and do
not include for-profit institutions. This set of schools includes schools
that do not have need-blind admission policies and therefore would not be
able to participate in activities allowed under the exemption.
13Comparable schools include the seven schools selected as control schools
for our econometric analysis.
Figure 2: Average Tuition, Fees, and Room and Board at Schools Using the
Antitrust Exemption Compared to All Other Private 4-Year Not-For-Profit
Schools and Comparable Schools, School Years 2000 to 2005
Note: Comparable schools include the seven schools selected as control
schools for our econometric analysis.
Percentage of Students Receiving Institutional Grant Aid and the Amount Schools
Provided Them Increased
Over the same time period, the percentage of students who received any
form of institutional grant aid at schools using the exemption increased
by 3 percentage points, from 37 to 40 percent, as illustrated by figure 3.
Figure 3: Percentage of Students at Schools Using the Antitrust Exemption
Receiving Various Types of Institutional Grant Aid from 2000 to 2006
Note: Data collected from 26 of the 28 schools using the antitrust
exemption.
Among students receiving institutional grant aid, the percentage of
students receiving need-based grant aid increased from 34 to 36 percent
from 2000 to 2006. The percentage of students receiving non-need-based
grant aid also increased slightly, from 2 to 4 percent. Non-need-based aid
is awarded based on a student's academic or athletic achievement and
includes fellowships, stipends, or scholarships. The majority of schools
using the exemption did not offer any non-need-based institutional grant
aid in school year 2005-2006. However, in 2005-2006 some schools did,
allocating non-need-based grant aid to between 16 to 54 percent of their
students.
As the cost of attendance and percentage of students receiving
institutional aid rose, participating institutions increased the amount of
such aid they provided students, although the percentage increases in aid
were smaller. As shown in figure 4, the average need-based grant aid award
across the schools using the exemption increased from $18,925 to $20,059,
or 6 percent. The average amount of non-need-based grant aid awards
dropped slightly from $12,760 in 2000-01 to $12,520 in 2005-06, or 2
percent. Overall, the average total institutional grant aid awarded to
students, which included both need and non-need-based aid, increased from
$18,675 in 2000-01 to $19,901 in 2005-06, or 7 percent.
Figure 4: Average Amount of Various Institutional Grant Aid Awards at
Schools Using the Antitrust Exemption from 2000 to 2006
Note: Data collected from 26 of the 28 schools using the exemption.
Students Accepted to Both Schools Using the Exemption and Comparable Schools Had
No Appreciable Difference in the Amount They Would Be Expected to Contribute
Towards College
There was virtually no difference in the amounts students and their
families were expected to pay between schools using the exemption and
similar schools not using the exemption. Average EFC was $27,166 for
students accepted at schools using the exemption, and $27,395 for those
accepted at comparable schools not using the exemption in school year
2005-2006. Moreover, the variation in the EFC for a student who was
accepted to several schools using the exemption was similar to the
variation in EFC that same student received from schools not using the
exemption. The variation in EFCs for these students was about $6,000 at
both sets of schools.14 Because the number of such students was small, we
also analyzed variation in EFCs for students who were accepted only at
schools using the exemption and compared it to the variation for students
who were only accepted at comparable schools not using the exemption.15 We
found slightly greater variation among EFCs for students who were accepted
at schools using the exemption; however, because we could not control for
student characteristics, factors external to the exemption could explain
this result, such as differences in a family's income or assets.
Although officials from schools using the exemption expected that students
accepted at several of those schools would experience less variation in
the amounts they were expected to pay, none of our analyses confirmed
this. The lack of consistency in EFCs among schools using the exemption
may be explained by the varied implementation of the consensus approach.
As previously mentioned, not all schools using the consensus approach
chose to adopt all the elements of the methodology. For example, seven
schools chose not to use the consensus approach to home equity, which uses
a percentage of the home equity in calculating the EFC. Using another
method for assessing a family's home equity could significantly affect a
student's EFC. For instance, we estimated that a family residing in
Maryland with an income of $120,000 and $350,000 in home equity would have
an EFC of $58,243 if a school chose not to implement the home equity
option in the consensus approach. Under the consensus approach, the amount
of home equity included in asset calculations would be capped and only
$38,000 of the home's equity would be included in the calculation of EFC.
The same family would then have an EFC of $42,449 if the school chose to
implement the option.
14Variation was measured by the standard deviation of the EFCs.
15For a more detailed discussion of our analysis see appendix I.
Implementation of a Common Methodology Has Not Significantly Affected
Affordability or Enrollment at Schools Using the Exemption
Based on our econometric analysis, schools' use of the consensus approach
did not have a significant impact on affordability, nor did it cause
significant changes in the likelihood of student enrollment at schools
using the consensus approach compared to schools that were not using the
consensus approach.16 As shown in table 5, while we found that the
consensus approach resulted in higher need-based grant aid awards for some
student groups (middle income, Asian students, and Hispanic students)
compared to similar students at schools that were not using the consensus
approach, this increase was likely offset by decreases in non-need-based
grant aid, such as academic or athletic scholarships.17 Thus, total grant
aid awarded was not affected by the consensus approach because the
increase in need-based aid was likely offset by decreases in
non-need-based grant aid.18
16GAO's econometric analysis was focused on the mandate from Congress that
requires us to examine the effects of the exemption. It is different from
a market-specific analysis conducted in an antitrust investigation, and is
not intended to address whether or not conduct may be taking place that
might violate the antitrust laws in the absence of the exemption.
17The results were similar for need-based institutional grant aid.
18The discussed effects of the consensus approach are statistically
significant (i.e., different from zero) at the 5 percent significance
level or less.
Table 5: Estimated Changes in Amount Paid, Financial Aid, and Enrollment
at Schools Using the Consensus Approach Compared to Schools Not Using the
Exemption
Estimated changes of using the consensus
approach on:
Amount Total Need-based Probability
students grant total grant Total aid (grant, of
Student group paid aid aid loans, work-study) enrollment
All students $6,125b
[$239,
$3,021 -$749 $12,011] -$2,886 38%
Financial-aid
applicants 2,177 n/a n/a n/a 22
Low-income 12,121b
-4,061 3,688 1,956 [1,837, 22,404] 59
Lower-middle
income 8,089 c -3,671 6,556 -7,776 95
Middle income 20,221a
[6,718,
2,320 1,618 33,724] 1,178 26
Upper-middle
income -1,048 -973 2,769 -3,054 18
High income 3,699 -714 4,687c -3,856 31
Asian students 14,628a
[5,051,
-376 5,726 24,206] 3,694 1
Black students 4,468 -1,227 4,332 -6,542 -26
Hispanic 9,532b
students
[1,006,
1,168 1,520 18,059] 3,648 108
White students 6,017b
[178,
2,588 -491 11,856] -2,879 19
Source: GAO analysis (see table 16 in app. II).
aResult is statistically significant at the 1 percent level or lower.
bResult is statistically significant at the 5 percent level or lower.
cResult is statistically significant at the 10 percent level or lower.
Notes: The estimates in brackets are the confidence levels of the
estimates that are significant at the 5 percent or lower level.
n/a means not applicable because of data limitations.
All the monetary values are in 2005 dollars.
Amount students paid is defined as tuition, room, board, fees, and other
expenses minus grant aid.
Total grant aid includes both need- and non-need-based aid from federal,
state, institutional and other sources.
Total aid includes grants, loans, and work-study aid from federal, state,
institutional, and other sources.
The effect of the consensus approach on need-based institutional grant aid
was $6,020, significant at the 5 percent level, with confidence interval
between $512 and $11,528.
The value of the effect of the consensus approach on institutional grant
aid was $1,331, but not statistically significant.
A different effect was found when low-income students at schools using the
consensus approach were compared to their counterparts at schools not
using the consensus approach. As shown in table 5, low income students at
schools using the consensus approach received, on average, a significantly
higher amount of total aid-about $12,121, which includes both grants and
loans. However, the amount of grant aid that these students received did
not significantly change, suggesting that that they likely received more
aid in the form of loans, which would need to be repaid, or work-study.
Our analysis of the effects of the consensus approach on various racial
groups showed no effect on affordability for these groups compared to
their counterparts at schools not using the consensus approach. While
Asian, white, and Hispanic students received more need-based grant aid
compared to their counterparts at schools not using the consensus
approach, their overall grant aid awards did not change.
Finally, as shown in table 5, there were no statistically significant
effects of the consensus approach on student enrollment compared to the
enrollment of students at schools not using the consensus approach. In
particular, the consensus approach did not significantly increase the
likelihood of enrollment of low-income or minority students or any student
group.
Our econometric analysis has some limitations that could have affected our
findings.19 For example, we could not include all the schools using the
consensus approach in our analysis because there were no data available
for some of them. However, there were enough similarities (in terms of
"best college" ranking, endowment, tuition and fees, and percentage of
tenured faculty) between the included and excluded participating schools
that allowed for a meaningful analysis. (See table 6 for a list of schools
included in our analysis).
19For a more detailed discussion on our econometric models and the
limitations of our analysis see appendix II.
Table 6: Schools Included in Analysis of Effects of Exemption
Comparable schools not using the
Schools using the consensus approach consensus approach
Cornell University Duke University Brandeis University Bryn Mawr College
Georgetown University University of New York University Princeton
Notre Dame Vanderbilt University University Tulane University
Wake Forest University Yale University of Rochester Washington
University University at St. Louis
Source: GAO analysis.
Moreover, the data for our post-consensus approach period was collected in
2003-2004-the first or second year that some schools were using the
consensus approach. Because we have data for only one year after
implementation, it is possible that some eventual effects of the consensus
approach may not be captured. The effects of using the consensus approach
could be gradual, rather than immediate, and therefore may not be captured
until later years.
Concluding Observations
By providing an exemption to antitrust laws enabling schools to
collaborate on financial aid policies, the Congress hoped that schools
would better target aid, making college more affordable for low income and
other underrepresented groups. The exemption has not yet yielded these
outcomes. Nor did our analysis find an increase in prices that some feared
would result from increased collaboration among schools. Initial
implementation of the approach has been varied; some schools have not
fully implemented the need analysis methodology, and many schools are
still in the initial years of implementation. As is often the case with
new approaches, it may be too soon to fully assess the outcomes from this
collaboration.
Agency Comments
We provided the group of schools using the antitrust exemption, the
Secretary of Education, and the Attorney General with a copy of our draft
report for review and comments. The group of schools using the exemption
provided written comments, which appear in appendix IV. In general, the
group stated that our study was a careful and objective report, but raised
some concerns about the data used in our econometric analysis and the
report's tone and premise. Specifically, they raised concerns about the
selection of treatment and control schools for our econometric analysis.
As we noted in the report, we selected schools for selection in treatment
and control groups based, in part, on the availability of student-level
data in the NPSAS. Some schools that used the consensus approach were not
included because there were no data available for them. However, we
believe there were enough similarities between the included and excluded
schools to allow for a meaningful analysis. The group also stated that a
number of conclusions were based on a very small number of observations.
In appendix II, we acknowledge the small sample size of the data could
make the estimates less precise, especially for some of the subgroups of
students we considered. However, we performed checks to ensure that our
estimates were reliable and believe that we can draw conclusions from our
analysis. With respect to the tone and premise of the report, the group
raised concerns about using low income students as "a yardstick for
judging the success of the Consensus Approach." When passing the
exemption, Congress hoped that it would further the government's goal of
promoting equal access to educational opportunities for students.
Need-based grant aid is one way to make college more affordable for the
neediest students to help them access a post-secondary education. The
group also highlighted several positive outcomes from their collaboration,
including a more transparent aid system and more engagement by college
presidents in aid-related discussions, topics which our study was not
designed to address. The group provided technical comments, which we
incorporated where appropriate. Education reviewed the report and did not
have any comments. The Department of Justice provided technical comments,
which we incorporated where appropriate.
We are sending copies of this report to the Secretary of Education,
Attorney General, appropriate congressional committees, and other
interested parties. In addition, the report will be available at no charge
on GAO's Web site at http://www.gao.gov .
If you or your staff have any questions please call me on (202) 512-7215.
Contact points for our Offices of Congressional Relations and Public
Affairs may be found on the last page of this report. Other contacts and
staff acknowledgments are listed in appendix VI.
Cornelia M. Ashby, Director Education, Workforce and Income Security
Issues
List of Congressional Committees
The Honorable Arlen Specter Chairman
The Honorable Patrick J. Leahy Ranking Minority Member Committee on the
Judiciary United States Senate
The Honorable F. James Sensenbrenner, Jr. Chairman
The Honorable John Conyers, Jr. Ranking Minority Member Committee on the
Judiciary House of Representatives
Appendix I: Statistical Analysis of Expected Family Contributions at
Schools Using the Exemption and Comparable Schools
We compared variation in expected family contributions (EFCs) between
students who were admitted to both schools using the exemption and
comparable schools that did not. We collected data on student EFCs from 27
of the 28 schools using the exemption and 55 schools that had similar
selectivity and rankings as schools using the exemption. The data included
the student's EFC calculated by the schools as of April 1, 2006, based on
their need analysis methodology. We determined that these data would most
likely reflect the school's first EFC determination for a student and thus
would be best for comparison purposes. We then matched students across
both sets of schools to identify students accepted to more than one school
(which we call cross-admits).
Our sample consisted of data for the following three types of cross-admit
students:
1. Students accepted to several schools using the exemption and
several schools that were not (type 1 students);
2. Students accepted to only schools using the exemption (type 2
students); and
3. Students accepted to only schools not using the exemption (type
3 students).
Data from the type 1 sample provided the most suitable data for our
analysis because it controlled for student characteristics. However,
because this sample was relatively small, we used the other samples to
supplement the analysis.
Once the cross-admits were identified, the EFCs for each student were used
to evaluate the mean and median as measures of location and the standard
deviation and range as measures of variation. Given the potential scale
factor, the variation measures were standardized. The standard deviation
was standardized by dividing it by the mean, and the range was
standardized by dividing it by the median. The two resulting variation
measures were the coefficient of variation (V1) and its robust counterpart
(V2), respectively.
These two measures of variation were estimated for each and every student.
The estimates were grouped for both sets of schools. We labeled schools
using the exemption as "568 schools" and comparable schools that were not
as "non-568 schools."
Table 7 reports various estimates averaged over students in each group.
The table generally shows similar group averages for the mean, standard
deviation, median, and range that were used to compute V1 and V2. The
values reported are the averages for all the students in each group. There
are fewer observations for the 568 schools than for the non-568 school,
except for type 1 students where the number of observations were equal
because the students were in both groups of colleges. In addition, we
imposed the following three conditions:
o First, for the coefficient of variation V1, we excluded all
observations where the standard deviations were zero. The zero
standard deviations are excluded because some of the non-568
schools that use only the federal methodology to calculate EFCs
report the same EFCs for a student and are likely to bias the
results. None of the observations with zero standard deviations
that we excluded involved a 568 school.
o Second, for the coefficient of variation V2, we excluded all
observations where the medians were zero because we could not
construct this measure that was obtained by dividing the range by
the median.
o And, third, for the coefficient of variation V2, we excluded
observations where the standardized variation exceeded 3 based on
the observed distributions of the data.
The test results were similar when none of those conditions were
imposed.
Table 7: Summary Statistics of Expected Family Contributions
Schools using the exemption Comparable schools
(568 schools) (Non-568 schools)
Type 1 Type 1
Students students All students students All students
Statistic
Standard deviation $6,188 $6,447 $6,190 $7,035
Mean $27,166 $31,640 $27,395 $28,747
[$22,576, [$30,380, [$22,293, [$27,924,
$31,757] $32,900] $32,497] $29,571]
Coefficient of
variation 1 (V1) 0.27 0.24 0.36 0.35
Range $12,200 $12,886 $9,671 $8,813
Median $30,374 $31,677 $29,225 $31,075
[$25,380, [$30,394, [$23,858, [$30,314,
$35,367] $32,961] $34,593] $31,836]
Coefficient of
variation 2 (V2) 0.47 0.49 0.52 0.37
Number of students N1=79 N1=1,158 N1=79 N1=2,866
N2=76 N2=1,150 N2=76 N2=3,653
Source: GAO analysis.
Notes: Coefficient of variation 1 (V1) equals standard deviation divided
by mean.
Coefficient of variation 2 (V2) equals range divided by median.
Type 1 consists of students with multiple offers from 568 colleges as well
as offers from non-568 colleges. For the 568 colleges, all students
consist of type 1 and type 2-students with multiple offers from only 568
colleges. And for the non568 colleges, all students consist of type 1 and
type 3-students with multiple offers from only non-568 colleges.
The values in brackets are the 95 percent lower and upper bounds
(confidence intervals).
N1 is the sample size for coefficient of variation 1 (V1) and N2 is sample
size for coefficient of variation 2 (V2).
Denoting the estimates of V1 and V2 for the two groups by and , and and ,
the empirical distribution of was then compared with the empirical
distribution of to examine whether and had identical distributions (that
is EFCs for 568 schools were similar in variations to those for non-568
schools). A similar comparison was made using the robust measures , and.1
To more closely examine the difference between the variations in EFCs of
cross-admit students for 568 and non-568 schools, we performed the
Kolmogorov-Smirnov test. The test examines whether the distributions of
the variation measures and were the same. The same analysis was done for
the V2 measures. The test was reported for both samples, consisting of
type 1 students and all students. The results reported in table 8 suggest
that there was no difference in EFC variations across the two groups,
using type 1 students. The results using all students, however, are
inconclusive for the V1 estimate, but suggest that non-568 schools have
smaller EFC variation for the V2 estimate. The results based on the type 1
sample are more useful as a stand-alone descriptive finding, because this
sample controls for student characteristics. The finding based on the
combined data requires further analysis to control for student
characteristics that we were unable to perform due to data limitations.
1We used the KSMIRNOV command in Stata to perform the tests.
Table 8: Tests of Variations in Expected Family Contributions
Alternative Test-statistic,
Variable Student Data hypothesis D p-value Conclusion
Coefficient Type 1 N1=79 Non-568 0.1013 0.445 EFCs are similar
of N2=79 EFCs are
variation 1 smaller -0.1519 0.162 EFCs are similar
(V1)
Non-568 Overall-EFCs are
EFCs are similar
larger
Coefficient Type 1 N1=76 Non-568 0.1447 0.203 EFCs are similar
of N2=76 EFCs are
variation 2 smaller -0.1053 0.431 EFCs are similar
(V2)
Non-568 Overall-EFCs are
EFCs are similar
larger
Coefficient All Non-568 0.1724 0.000 Non-568 EFCs are
of N1=1,158N2=2,866 EFCs are smaller
variation 1 smaller -0.1788 0.000
(V1) Non-568 EFCs are
Non-568 larger
EFCs are
larger Overall-Inconclusive
Coefficient All Non-568 0.3970 0.000 Non-568 EFCs are
of N1=1,150N2=3,653 EFCs are smaller
variation 2 smaller -0.0399 0.061
(V2) EFCs are similar
Non-568
EFCs are Overall-Non-568 EFCs
larger are smaller
Source: GAO analysis.
Notes: Coefficient of variation 1 (V1) equals standard deviation divided
by mean.
Coefficient of variation 2 (V2) equals range divided by median.
All means students with multiple offers from 568 schools as well as offers
from non-568 schools (type 1), students with multiple offers from only 568
schools (type 2), and students with multiple offers from only non-568
schools (type 3).
The p-values are for the Kolmogorov-Smirnov tests of equality of
distribution functions. All tests are interpreted using the 5 percent or
lower level of significance.
N1 is the sample size for coefficient of variation 1 (V1) and N2 is sample
size for coefficient of variation 2 (V2).
Appendix II: Econometric Analysis of Effects
of the Higher Education Antitrust Exemption on College Affordability and
Enrollment
To estimate the effects of schools' implementation of the consensus
approach to need analysis on affordability (measured by price) and
enrollment of freshmen students, we developed econometric models. This
appendix provides information on theories of the exemption effects on
student financial aid, the data sources for our analyses and selection of
control schools, specifications of econometric models and estimation
methodology, our econometric results, and limitations of our analysis.
Theories of the Effects of the Consensus Approach on Financial Aid
Two theories exist about the effects the consensus approach on student
financial aid. It is important to note that the award of grant aid
represents a discount from the nominal "list price", which lowers the
price students actually pay for college. So, any decision to limit grant
aid would be an agreement to limit discounts to the list price, and thus
may raise the price some students would pay. It is also important to note
that schools admit only a limited number of students. One of the theories
suggests that allowing schools a limited degree of collaboration could
reduce the variation in financial need determination for an individual
student and reduce price competition among colleges vying for the same
students. While the reduced competition would imply lower financial aid
(hence higher prices) for some students, schools could thus devote more
financial aid resources to providing access to other students, especially
disadvantaged students. This "social benefit theory" assumes that under
these conditions disadvantaged students would receive more grant aid and
as a result, pay less for school. Also, an implicit assumption of this
theory is that the exemption would essentially result in redistribution of
financial aid without necessarily changing the amount of financial aid
resources available. Moreover, because costs to students and their
families would change for some students, enrollment of such students would
be affected.
An opposing theory is that the exemption will allow schools to coordinate
on prices and reduce competition. This "anti-competitive theory"
essentially views coordination by the group as restraining competition.
Specifically, under this theory, allowing an exemption would result in
less grant aid and higher prices on average, especially for students that
schools competed over by offering discounts on the list price. As a
result, the amount of financial aid available to some students would
likely decrease. If prices are higher on average, it could cause a
decrease in enrollment, particularly of disadvantaged students since they
would be less able to afford the higher prices.1 Our analyses allowed us
to test these two theories with the data available.
Sources of Data for the Model
To construct our model, we used data from:
o National Postsecondary Student Aid Study (NPSAS): These data,
available at the student-level, served as the primary source for
our study because we were interested in student outcomes of the
exemption. Data were published every 4 years during the period
relevant to our study; hence, we have data for academic years
1995-1996, 1999-2000, and 2003-2004. The data contained
student-level information for all freshmen enrollees in the
database, including enrollment in school, cost of attendance,
financial aid, Scholastic Aptitude Test (SAT) scores, household
income, and race. The number of freshmen in the database for our
study was 1,626 in 1995-1996, 272 in 1999-2000, and 842 in
2003-2004.
o Integrated Postsecondary Education Data System (IPEDS): These
data, available at the school level, included tuition and fees,
faculty characteristics, and student enrollment for 1995-1996 and
2003-2004, there were no data published for 1999-2000. However,
some of the data for 1999-2000 were reported in the subsequent
publications. We were able to construct some data for 1999-2000
through linear interpolation of the data for 1998-1999 and
2000-2001 or using the data for either year depending on
availability; we believed this was reasonable because data for
these institutions did not vary much over time.2
o National Association of College and University Business
Officers (NACUBO): This source provided data on school endowment
from 1992 through 2004.3
o GAO Survey: The survey collected data on the activities of the
schools using the higher education antitrust exemption, including
when schools implemented the consensus approach methodology.
Selection of Control Schools
Determining the effects of the exemption required both a treatment
group (schools using the exemption) and a control group (a
comparable set of schools that did not use the exemption). To find
a comparable set of schools we used data on school rankings based
on their selectivity from years 1994 to 2004 from the U.S. News
and World Report (USNWR). We selected control schools similar to
schools using the antitrust exemption that had comparable student
selectivity and quality of education using the "best schools"
rankings information in the USNWR.4 The combined control and
treatment schools were matched to school-level data from IPEDS,
and student-level data from NPSAS. We selected the control schools
based on their ranks in the years prior to the implementation of
the consensus approach-1995-1996 and 1999-2000-and after the
implementation of the consensus approach-2003-2004. The USNWR
published its "best schools" rankings annually in August or
September. Thus, the 2004 publication reflected the selectivity of
the schools during 2003-2004. However, because publications in
prior years-2002 and 2003-provided relevant information to
students who enrolled in 2003-2004, we considered the rankings
published from 2002 through 2004 as important input into decisions
made by students and the schools for 2003-2004. Similarly, the
publications from 1994 through 1996 were used to determine the
selectivity of the schools in 1995-1996, and the publications from
1998 to 2000 were used to determine school selectivity for
1999-2000.
The USNWR published separate rankings for liberal arts schools and
national universities. The schools using or affiliated with the
exemption consisted of 28 current members, two former members, and
six observers.5 These 36 schools comprised the treatment schools
used initially to select the comparable control schools. All 36
treatment schools were private; 13 were liberal arts schools and
23 were national universities.6 To ensure there were enough
control schools for the treatment schools, we initially selected
all the schools ranked in tier 1 (and tier 2 when available) in
the USNWR rankings for each of the two types of
institutions-liberal arts schools and national universities.7 This
resulted in 250 schools, including all 36 treatment schools, for
nine selected years (1994 to 1996, 1998 to 2000, and 2002 to
2004). All the treatment schools were ranked in each of the nine
years (except for one school that was not ranked in 2002). The
initial list of 250 schools was refined further to ensure a proper
match in selectivity between the treatments and controls.
Although we were interested in obtaining an adequate number of
control schools to match the treatment schools, we refined the
selection process to ensure they were comparable using the
following conditions. First, we limited the selection of all the
schools (controls and treatments) to those that were ranked in
tier 1. This reduced the sample of schools from 250 to 106
schools, comprising all 36 treatment schools and 70 control
schools. Second, the list of 106 schools was used to match
school-level data from the IPEDS in each of the three academic
years.8 Third, these data were then matched with the IPEDS data
for each of the three academic years to student-level data from
NPSAS. From the NPSAS, we selected data for cohorts who entered
their freshmen year in each of the three academic years.9 Fourth,
since we used a difference-in-difference methodology for the
analysis, we wanted data for each school in at least two of the
three academic years-one in the pre-treatment and one in the
post-treatment period. We therefore initially constructed four
samples of schools, depending on whether there were matches
between all three academic years, or between any two of the three
academic years. This resulted in 30 schools with data in all three
academic years 1995-1996, 1999-2000, and 2003-2004 (referred to as
sample 1). There were 34 schools with data in 1995-1996 and
2003-2004 (sample 2); 35 schools with data in 1999-2000 and
2003-2004 (sample 3); and 37 schools matched between 1995-1996 and
1999-2000 (sample 4).10 Finally, we limited the selection to
private schools because all of the treatment schools are private.
We did this because the governance of private schools generally
differed from state-controlled public schools and these
differences were likely to affect affordability and enrollment at
a school.
Determination of the Appropriate Time Periods for Assessing Effects
and Classification of Schools that Only Attended the Meetings
We also determined the academic year(s) data that would be used to
represent the period before and the period after the
implementation of the consensus approach. Since we had data for
only1995-1996, 1999-2000 and 2003-2004, and given that the
consensus approach was implemented in 2003-2004 (or in the prior
year by some schools) we selected 1995-1996 as the pre-consensus
approach period and 2003-2004 as the post-consensus approach
period. Although the 1999-2000 data were relatively current for
the pre-consensus approach period, it is possible that the
1999-2000 data may offer neither strong pre- nor post-consensus
approach information since the period was very close to the
formation of the 568 President's Group in 1998. Furthermore, the
institutional methodology, which is a foundation for the consensus
approach and used by some of the control schools in 2003-2004, was
revised in 1999. We therefore investigated whether it was
appropriate to include 1999-2000 in the pre-consensus approach
period or in the post-consensus approach period. We also
investigated in which group (control or treatment) the schools
that only attended the 568 President's Group meetings, but had not
become members of the group or implemented the consensus approach,
belonged.
Using the Chow test for pooling data, we determined that 1999-2000
should be excluded from the pre-consensus approach period as well
as from the post-consensus approach period. We also determined
that schools that only attended the 568 President's Group meetings
could not be regarded as control schools or treatment schools in
analyzing the effects of the consensus approach.11 Therefore, the
treatment schools consisted of the group members that implemented
the consensus approach, and the control schools consisted of the
schools that were not members of the 568 Group and did not attend
their meetings. Based on the analysis above, we used the data in
sample 2, which excluded data collected in 1999-2000, for our
baseline model analysis; the period before the consensus approach
is 1995-1996 and the period after is 2003-2004; the control
schools that did not use the consensus approach (non-CA schools)
are Brandeis University, Bryn Mawr College, New York University,
Princeton University, Tulane University, University of Rochester,
and Washington University at St. Louis, and the treatment schools
that used the consensus approach (CA schools) are Cornell
University, Duke University, Georgetown University, University of
Notre Dame, Vanderbilt University, Wake Forest University, and
Yale University. The complete list of the schools is in table 9.
1This theory is consistent with the idea that non-profit organizations
have an incentive to exercise market power despite not directly capturing
profits, because the extra resources from exercising market power allow
them to invest in other areas they deem important; e.g., schools may
charge high prices to students because it could enable them to offer
higher salaries to attract high-caliber faculty.
2Student enrollment data was obtained through linear interpolation, and
faculty data was based on 1998-1999 data.
3Where necessary, the data were supplemented by data from IPEDS.
Selection of Control Schools
4Schools were ranked annually based on various criteria (including
selectivity, faculty and financial resources, graduation rate, and alumni
satisfaction) in various publications-particularly in the USNWR, the
Peterson's Four-Year Schools, and the Barron's Profiles of American
Schools. The rankings of the schools by the different publishers were
generally similar, but since the data were readily available in the USNWR
we chose its rankings. Using the published rankings helped avoid a
possible bias from arbitrarily picking the schools. Furthermore, these
rankings were widely used and generally stable over time.
5Although the observers were not members they attended the group's
meetings. The former members were Bowdoin College and Macalester College.
6Liberal arts schools emphasize undergraduate education and award at least
half of their degrees in the liberal arts discipline, and most are
private. National universities offer a wide range of undergraduate majors
as well as master's and doctoral degrees, and many emphasize research.
7The number of schools in the two tiers for each type of school was
between 50 and 90 for each year.
8We also used endowment data from NACUBO, and school-level data from GAO's
survey of the schools.
9We used data for students who were enrolled as freshmen, as of October of
the academic year, in the NPSAS database.
Determination of the Appropriate Time Periods for Assessing Effects and
Classification of Schools that Only Attended the Meetings
10Although the sample periods used by Hoxby (2000) and Netz (2000) are
much earlier than what we used, our list of schools is reasonably
consistent with theirs. Similarly, our list of schools was consistent with
the schools in the Consortium for Financing Higher Education (COFHE),
which are some of the most selective private schools in the U.S.
11See appendix III for details of the tests.
Table 9: Control and Treatment Schools for Analyzing Effects of the
Consensus Approach Implementation
Academic years Control school (Non-CA) Treatment school (CA)
1995-1996 Sample 1: Brandeis Samples 1, 2, or 3: Boston
1999-2000 & University New York Collegea Cornell Universityb Duke
University Princeton University Georgetown University
2003-2004 Universityb Tufts Massachusetts Institute of
Universitya,b Tulane Technologya,b University of Notre
University University of Dame University of
Rochester Washington Pennsylvaniaa,b Vanderbilt
University at St. Louis University Wake Forest University
1995-1996 & Sample 2-All of Sample 1 Yale Universityb
Plus: Bryn Mawr Collegeb
2003-2004 Yeshiva Universitya
1999-2000 & Sample 3-All of Sample 1
Plus: Colgate University
2003-2004 Lehigh University Whitman
College
1995-1996 & Sample 4-All of Sample 1 Sample 4-All of Above Plus:
Plus: Carnegie Mellon Columbia Universityb
1999-2000 University Johns Hopkins
University
Source: GAO analysis.
aSchools were excluded because there were no data for SAT scores for
2003-2004.
bMember of the former Overlap group.
cMembers of the 568 Group that had not implemented the consensus approach.
dWere not members of the 568 Group but attended meetings.
eFormer member of the 568 Group.
Notes: Schools that Only Attended 568 Group Meetings: Sample 2: Stanford
Universityd, University of Southern Californiaa,d and Sample 4: Case
Western Reserve Universitye .
Member schools that had not implemented the consensus approach: Sample 2:
Brown University,b,c Sample 3: Dartmouth Collegeb,c.
Other 568-Affiliated Schools: Amherst College,b Bowdoin College,b,e
California Institute of Technology,d Claremont McKenna College, Davidson
College, Emory University, Grinnell College, Harvard University,b,d
Haverford College, Macalester College,e Middlebury College,b Northwestern
University, Pomona College, Rice University, Swarthmore College, Syracuse
University,d University of Chicago, Wellesley College,b Wesleyan
University,b Williams Collegeb
Specifications of Econometric Models and Estimation Methodology
We developed models for analyzing the effects of the implementation of the
consensus approach (CA) on affordability and enrollment of incoming
freshman using the consensus approach.12 We used a
difference-in-difference approach to identify the effects of
implementation of the consensus approach. This approach controlled for two
potential sources of changes in school practices that were independent of
the consensus approach. First, this approach enabled us to control for
variation in the actions of schools over time that were independent of the
consensus approach. Having control schools that never implemented the
consensus approach allowed us to isolate the effects of the exemption and
permitted us to estimate changes over time that were independent of the
consensus approach implementation. Second, while we had a control group of
schools that did not use the consensus approach, but were otherwise very
similar to treatment schools, it is possible that schools using the
consensus approach differed in ways that would make them more likely to
implement practices that are different from those of other schools.13 The
difference-in-difference approach controlled for this possibility by
including data on schools using the consensus approach both before and
after its adoption. Controlling then for time effects independent of the
consensus approach as well as practices by these schools before adoption,
the effect of the use of the consensus approach could be estimated.
Compared to the schools that did not use the consensus approach, we
expected that the implementation of the consensus approach would have a
significantly greater impact on the schools using the consensus approach
because its use has potential implications for affordability and
enrollment of students in these schools.
12 We did not separate the effects of the CA into the effects of only
attending the 568 Group meetings and the effects of only implementing the
CA, although some schools only attended meetings and had not implemented
the CA, because in table 9 there are only three schools in sample 2 that
would serve as treatments or serve as controls if we investigate the
effects of only attending meetings or the effects of only implementing the
CA, respectively.
13In addition to having the control schools, we also controlled for a
number of school characteristics that are discussed below. It is only the
possibility of changes in differences between treatment and control
schools that were not measurable or not observable that might lead to bias
in estimating the effects of the consensus approach implementation. For
example, schools adopting the consensus approach might differ in their
objectives concerning their preferred student body. As discussed next in
the text, the difference-in-difference approach provided controls for such
possibilities.
Modeling the Effects of the Consensus Approach Methodology for Financial Need on
Affordability and Enrollment
The basic tenets of financial need analysis are that parents and students
should contribute to the student's education according to their ability to
pay. The CA schools used the consensus approach for its need analysis
methodology and to determine the expected family contribution (EFC) for
each student based on that methodology. Conversely, the non-CA schools
primarily used a need analysis methodology called the institutional
methodology (IM). The difference between the cost of attendance (COA) and
the EFC determines whether a student has financial need. If so, the school
then develops a financial aid package of grants, loans, and work study
from various sources. The actual amount that students and families pay
depends on how much of the aid received is grant aid. Therefore, the
implementation of the consensus approach was expected to affect the price
paid and the financial aid received by students, and by implication, their
enrollment into schools.
Dependent variables:
The study examined the effects of the implementation of the consensus
approach on two key variables: affordability (measured by price) and
enrollment of freshman. We also estimated other equations to provide
further insights on affordability- tuition, total grant aid, need-based
grant aid, and total aid. All the dependent variables were measured at the
student level, except tuition. Also, all monetary values were adjusted for
inflation using the consumer price index (CPI) in 2005 prices.14 The
dependent variables were defined as follows:
o Price (PRICEijt): Price, in dollars, actually paid by freshman
i who enrolled in school j in an academic year t. The variable was
measured as the cost of attendance less total grant aid. The cost
of attendance consisted of tuition and fees, on-campus room and
board, books and supplies, and other expenses such as
transportation. Total grant aid consisted of institutional and
non-institutional grant aid; it excluded self-help aid (loans and
work study).
The other dependent variables that we estimated to help provide
more insights into the results for affordability were:
o Tuition (TUITIONijt): The amount of tuition and
fees in dollars charged by school j to freshman i who
enrolled in an academic year t.15
o Total grant aid (AIDTGRTijt): The amount of total
grant aid received, in dollars, by a freshman i who
enrolled in school j in an academic year t. The
counterpart to grant aid was self-help aid.16
o Need-based grant aid (AIDNDTGRTijt): The amount of
need-based grant aid received, in dollars, by
freshman i who enrolled in school j in an academic
year t. The counterpart to need-based aid was
non-need-based aid, which consisted mainly of merit
aid.17
o Total aid package (AIDTOTAMTijt): The amount of
total aid received, in dollars, by freshman i who
enrolled in school j in an academic year t. The total
aid consisted of total grants (from the school, the
various levels of government-federal, state-and other
sources) and self-help (includes loans and
work-study).
o Student enrollment (ENRCAijt): An indicator variable for
student enrollment into a CA school (ENRCAijt). It equals one if a
freshman i enrolled in an academic year t in school j that was a
school using or later the consensus approach, and zero otherwise.
Thus, at t=0 (1995-1996), a school was designated as a CA school
if it implemented the consensus approach in period t=1
(2003-2004). Students who enrolled in a non-CA school were
assigned a value of zero. In other words, ENRCA takes a value of
one for every student enrolled in a CA school in any time period
(1995-1996 or 2003-2004), and zero otherwise.
Explanatory variables:
Several variables could potentially affect each of the dependent
variables identified above. The explanatory variables we used were
based on economic reasoning, previous studies, and data
availability.18 All the equations used were in quasi reduced-form
specifications. The key explanatory variable of interest was the
exercise of the exemption through the implementation of the
consensus approach by the 568 Group of schools. We were also
interested in the effects of the implementation of the consensus
approach on affordability and enrollment of disadvantaged
students. In order to isolate the relationships between the
consensus approach implementation and each of the dependent
variables, we controlled for the potential effects of other
explanatory variables. The following is a complete list of all the
explanatory variables we used:
o Exemption indicator: EMCAjt 19
The exemption was captured by the implementation of the consensus
approach by a school.20 EMCA equals one if school j has
implemented CA by academic year t, where t is 2003-2004 and zero
otherwise.
We used other explanatory variables in our equations, in addition
to the exemption indicator for the implementation of the consensus
approach. These variables included school-level characteristics,
school specific fixed-effects, time specific fixed-effects, and
student-level characteristics.
o School-level characteristics:21
The school variables or attributes varied across the schools (j)
and over time (t), but did not vary across the students (i). The
school characteristics may capture the quality of the schools,
expenditures by the schools that may compete with financial aid
for funding, revenue sources for financial aid, or the preferences
of the students.22 The variables used were:
o ENDOWSTUjt: The interaction between the 3-year
average endowment per student and the 3-year average
percentage rate of return on endowment per student at
school j for an academic year t. The inclusion of the
rate of returns from endowments helped minimize the
possibility that developments in financial markets
could bias the results especially if the average
endowment per student differed across the two groups
of schools.
o RANKAVGjt: The average "best schools" rank of
school j for an academic year t. Although we used
this variable to select the control schools that were
comparable in selectivity to the treatment schools
before matching the data to the NPSAS data, this
variable was included, due to data limitations, to
control for the possibility that the two groups of
schools used in the sample may differ in selectivity.
o ENROLUGjt: The 3-year average growth rate (in
decimals) of undergraduate enrollment at school j for
an academic year t.
o TENUREDjt: The percentage (in decimals) of total
faculty at school j that was tenured in an academic
year t.
o Time specific fixed-effects:
These variables captured differences over time that did not vary
across the schools, such as increases in national income that
could increase affordability of schools. This was an indicator
variable for the academic years (time):
AY1995t: Equals one for the academic year 1995-1996, and zero
otherwise AY2003t: Equals one for the academic year 2003-2004, and
zero otherwise.
o Student characteristics:23
All the student-level variables or attributes generally varied
across students (i), across schools (j), and across time (t). The
student characteristics indicated the preferences of the students
for a school as well as the decisions of the schools regarding the
students they admitted. The variables used were:
o FINAIDijt: Equals one if a freshman i who enrolled
in school j in an academic year t applied for
financial aid, and zero otherwise.
o RACE: Equals one if a freshman i who enrolled in
school j in an academic year t is:
Asian-ASIANijt, and zero otherwise. Black-BLACKijt,
and zero otherwise. Hispanic-HISPANICijt, and zero
otherwise. White-WHITEijt, and zero otherwise.
Foreigner-FOREIGNijt, and zero otherwise. None of the
above-OTHERijt, and zero otherwise.24
o INCOME: Equals one for a freshman i who enrolled
in school j in an academic year t has household
income in the following quintiles:
INCLOijt: Below or equal to the 20th percentile, and
zero otherwise. These were low-income students, and
the median income for the group was $13,731 in 2005
dollars.
INCLOMDijt: Above the 20th and below or equal to the
40th percentile, and zero otherwise. These were
lower-middle income students, and the median income
for the group was $40,498 in 2005 dollars.
INCMDijt: Above the 40th and below or equal to the
60th percentile, and zero otherwise. These were
middle-income students, and the median income for the
group was $59,739 in 2005 dollars.
INCUPMDijt: Above the 60th and below or equal to the
80th percentile, and zero otherwise. These were
upper-middle income students, and the median income
for the group was $88,090 in 2005 dollars.
INCHIijt: Above the 80th percentile, and zero
otherwise. These were high-income students, and the
median income for the group was $145,912 in 2005
dollars.
Since we included minority students (Asian, black, and Hispanic
students) as well as lower income groups (low income and
lower-middle income students) to measure needy students, the
minority variables likely captured nonincome effects.25
EFCijt: Expected family contribution for a freshman i who enrolled
in school j in an academic year t. Although this variable captured
the income of the students, it also reflected other factors that
affect financial aid, such as the number of siblings in college.26
SCORESATijt: The combined scholastic aptitude test (SAT) scores
for math and verbal of freshman i who enrolled in school j in an
academic year t.
Tables 10 and 11 show summary statistics for the variables listed
above for treatment and control schools in sample 2 (as listed in
table 9).27 In general, the values of the variables were similar
between the two groups of schools.
14All the dependent variables were from NPSAS, except tuition, which was
from IPEDS. We used the general price level instead of the price index for
higher education to adjust the monetary values because the former better
reflected potential substitution effects between college education and
other expenditures by households. Furthermore, sector-specific price
indexes generally tend to be more volatile.
15The tuition amount was the same for all freshmen in a private school.
16We also estimated an equation for institutional grant aid (AIDINSTGRT)
and self-help aid (AIDSELFPLUS).
17We also estimated an equation for need-based institutional aid
(AIDNDINST) and non-need-based grant aid (AIDNONDTGRT), which was the
difference between total grant aid and need-based aid. However, we did not
have enough data to estimate merit-only aid.
18We relied on several previous studies, including Avery and Hoxby (2003),
Carlton et al. (1995), Bamberger and Carlton (1993), Epple et al. (2005),
Hill et al. (2005), Hoxby (2000), Kim (2005), Netz (1999, 2000), Hill and
Winston (2001), Morrison (1992), Salop and White (1991), Shepherd (1995),
and Winston and Hill (2005).
19This variable was from the GAO survey of the CA and non-CA schools.
20The CA schools are the 568 schools that have either implemented the
consensus approach fully or in part by implementing some of the options
under that need analysis methodology for financial aid. Of the seven CA
schools in sample 2 in table 9, only three had not fully implemented the
consensus approach (Georgetown, Vanderbilt, and Wake Forest).
21All the school-level variables are from IPEDS.
22The school specific fixed-effects were estimated using the fixed-effects
estimator, where feasible. This effect captured differences among the
schools that did not vary over time, such as location, memberships in
athletic conferences and other organizations such as the former Overlap
group. Also, several school-level variables could not be used in the
models because the variables did not vary over time, and were therefore
expected to be captured by the school specific fixed-effects.
23All the student-level variables were from NPSAS.
24We included Native Americans in OTHER because of their relatively small
numbers.
25To avoid the dummy-variable trap in the estimation, we excluded white
students from the racial groups, and high-income students from the income
groups.
26The EFC is the federal calculation, which differs significantly from the
EFC calculated by the CA schools, and to some extent from the EFC
calculated by the non-CA schools. We found a negative relationship between
the number of siblings and EFC using the limited data on siblings,
although the link was not strong.
27The reported values are probability-weighted.
Table 10: Summary Statistics of Variables Used in Regression Analysis,
1995-1996 and 2003-2004: CA Schools
Variable Mean Std Min Max
School-level
TUITION $26,245 $3,557 $18,910 $31,152
ENDOWSTU $227,213 $230,768 $44,061 $1,146,129
RANKAVG 16 9 2 27
ENROLUG 2% 6% -1% 21%
TENURED 56% 13% 25% 75%
Student-level
PRICE $30,792 $11,144 $1,065 $52,354
AIDTGRT $7,133 $9,866 $0 $40,658
AIDNDTGRT $5,526 $8,722 $0 $35,321
AIDNONDTGRT $1,607 $4,360 $0 $30,403
AIDTOTAMT $12,465 $13,566 $0 $43,195
AIDSELFPLUS $4,794 $8,155 $0 $36,730
EFC $24,486 $22,268 $0 $115,090
SCORESAT 1301 144 790 1600
FINAID 76% n/a n/a n/a
ASIAN 9% n/a n/a n/a
BLACK 5% n/a n/a n/a
HISPANIC 7% n/a n/a n/a
FOREIGN 2% n/a n/a n/a
OTHER 5% n/a n/a n/a
WHITE 71% n/a n/a n/a
INCLO 5% n/a n/a n/a
INCLOMD 11% n/a n/a n/a
INCMD 13% n/a n/a n/a
INCUPMD 17% n/a n/a n/a
INCHI 54% n/a n/a n/a
Schools Cornell University, Duke University, Georgetown University,
University of Notre Dame, Vanderbilt University, Wake Forest
University, Yale University
Number of
observations 241
Source: GAO analysis.
Note: All values are (probability) weighted averages, and the monetary
values are in 2005 dollars.
Table 11: Summary Statistics of Variables Used in Regression Analysis,
1995-1996 and 2003-2004: Non-CA Schools
Variable Mean Std Min Max
School-level
TUITION $27,031 $2,259 $24,571 $31,714
ENDOWSTU $256,147 $329,513 $27,909 $1,504,930
RANKAVG 23 12 1 43
ENROLUG 1% 1% -2% 4%
TENURED 56% 12% 26% 72%
Student-level
PRICE $28,815 $10,305 $4,569 $50,726
AIDTGRT $10,869 $9,792 $0 $32,803
AIDNDTGRT $8,573 $9,132 $0 $31,487
AIDNDTGRT $2,296 $5,419 $0 $27,919
AIDTOTAMT $16,487 $13,875 $0 $48,572
AIDSELFPLUS $5,293 $7,686 $0 $48,041
EFC $21,717 $21,724 $0 $105,095
SCORESAT 1268 151 740 1590
FINAID 80% n/a n/a n/a
ASIAN 12% n/a n/a n/a
BLACK 5% n/a n/a n/a
HISPANIC 4% n/a n/a n/a
FOREIGN 3% n/a n/a n/a
OTHER 3% n/a n/a n/a
WHITE 74% n/a n/a n/a
INCLO 10% n/a n/a n/a
INCLOMD 9% n/a n/a n/a
INCMD 12% n/a n/a n/a
INCUPMD 21% n/a n/a n/a
INCHI 48% n/a n/a n/a
Schools Brandeis University, Bryn Mawr College, New York University,
Princeton University, Tulane University, University of
Rochester, Washington University at St. Louis
Number of
Observations 277
Source: GAO analysis.
Note: All values are (probability) weighted averages, and the monetary
values are in 2005 dollars.
Comparison of Prices and Financial Aid in CA and Non-CA Schools
Table 12 shows summary statistics on price and financial aid before and
after the implementation of the consensus approach in 2003-04 at the CA
and non-CA schools in sample 2. Similarly, table 13 shows the summary
statistics by income and racial groups.28 It is important to note that the
summary information on the observed differences before and after the
implementation of the consensus approach for the CA and non-CA schools are
heuristic and do not conclusively determine the potential effects of the
implementation of the consensus approach. It is also important to note
that, for any given variable, it is possible that there are other factors
than implementing the consensus approach that are responsible for the
observed differences, including differences between CA and non-CA schools'
student populations or differences in the characteristics of the schools,
or both. For instance, the price paid by middle-income students increased
more in CA than in non-CA schools. While this may reflect the effect of
consensus approach, it is possible that other factors are responsible for
the differences. For example, the racial composition of middle-income
students might also be different between the two groups, or there may be
systematic differences in endowment growth between the CA and non-CA
schools that affect financial aid to middle-income students. Thus, to
assess the effect of consensus approach, it is necessary to study the
effects of consensus approach while controlling simultaneously for all
factors that influence price and aid policies.
28The reported values are probability-weighted.
Table 12: CA and Non-CA Schools: Price and Financial Aid
CA Schools Non-CA Schools
Percentage Percentage
All students 1995-1996 2003-2004 difference 1995-1996 2003-2004 difference
Pricea $28,039 $35,488 27% $28,068 $30,838 10%
Tuition &
fees 24,062 29,967 25 25,770 30,447 18
Total
observations 150 91 198 79
Financial-Aid Applicants Only
Student
Applied for
Financial Aid
Pricea $25,845 $32,897 27% $24,960 $29,705 19%
Total grant
aid 9,142 9,775 7 13,391 13,960 4
Need-based
total grant 7,771 6,439 -17 11,863 8,122 -32
Institutional
grant aid 7,073 6,529 -8 11,297 11,116 -2
Total aid 16,604 16,046 -3 19,827 22,255 12
Loans (incl.
PLUS) 5,954 4,849 -19 5,271 6,669 27
Work study 710 866 22 986 715 -27
Number of
observations 112 72 152 73
Student Did
Not Apply for
Financial
Aidb
Pricea 34,645 44,504 28 37,714 44,292 17
Number of
observations 38 19 46 6
Total
observations 150 91 198 79
Source: GAO analysis.
aPrice equals cost of attendance less total grant aid. Cost of attendance
equals tuition and fees, plus expenses (including room and board, and
books).
bFinancial aid data were not available for students who did not apply for
financial aid.
Notes: All values are (probability) weighted averages, and the monetary
values are in 2005 dollars.
Table 13: CA and Non-CA Schools-Financial Aid Applicants Only: Price and
Financial Aid
CA schools Non-CA schools
Percentage Percentage
1995-1996 2003-2004 difference 1995-1996 2003-2004 difference
Income level
Low income
Pricea $12,566 $10,095 -20 $18,950 $21,886 15
Total grant $19,093 $21,861 14
aid $23,429 $27,020 15
Need-based $16,849 $17,756 5
total grant $21,422 $21,191 -1
Institutional $14,060 $17,447 24
grant aid $17,278 $12,597 -27
Total aid $27,385 $35,956 31 $24,772 $26,523 7
Number of 26 8
observations 9 3
Lower-middle
income
Pricea $17,613 $30,437 73 $17,623 $20,546 17
Total grant $20,598 $23,014 12
aid $17,531 $14,793 -16
Need-based $20,417 $15,456 -24
total grant $15,762 $12,949 -18
Institutional $15,742 $19,386 23
grant aid $11,735 $9,667 -18
Total aid $24,025 $18,409 -23 $28,462 $30,509 7
Number of 22 5
observations 13 12
Middle income
Pricea $22,146 $30,156 36 $21,240 $25,279 19
Total grant $17,173 $17,336 1
aid $12,277 $12,076 -2
Need-based $16,053 $9,048 -44
total grant $8,293 $10,767 30
Institutional $16,000 $13,854 -13
grant aid $11,096 $9,936 -10
Total aid $18,811 $20,743 10 $24,261 $25,176 4
Number of 20 12
observations 16 10
Upper-middle
income
Pricea $23,759 $31,631 33 $26,905 $29,524 10
Total grant $12,030 $13,478 12
aid $10,410 $10,374 -0.3
Need-based $10,289 $6,593 -36
total grant $9,732 $5,864 -40
Institutional $10,900 $7,198 -34
grant aid $8,694 $7,719 -11
Total aid $16,926 $19,277 14 $19,425 $20,769 7
Number of 32 11
observations 21 13
High income
Pricea $31,776 $37,127 17 $30,184 $33,806 12
Total grant $7,941 $10,331 30
aid $3,493 $5,468 57
Need-based $6,185 $5,377 -13
total grant $2,810 $1,694 -40
Institutional $7,062 $8,884 26
grant aid $2,532 $3,430 35
Total aid $12,393 $10,850 -12 $13,300 $19,797 49
Number of 52 37
observations 53 34
Total 152 73
observations 112 72
Raceb
Asian
Pricea $28,082 $28,756 2 $25,642 $27,624 8
Total grant $10,827 $17,834 65
aid $8,371 $16,265 94
Need-based $9,900 $14,646 48
total grant $7,675 $13,129 71
Institutional $7,771 $13,376 72
grant aid $6,906 $11,607 68
Total aid $14,343 $23,037 61 $15,513 $27,425 77
Number of 23 13
observations 11 5
Black
Pricea $12,702 $22,935 81 $13,530 $17,375 28
Total grant $23,296 $25,010 7
aid $21,360 $19,958 -7
Need-based $18,517 $15,631 -16
total grant $19,836 $8,932 -55
Institutional $18,582 $21,231 14
grant aid $15,046 $17,404 16
Total aid $29,572 $24,950 -16 $29,121 $26,707 -8
Number of 8 4
observations 10 3
Hispanic
Pricea $21,177 $21,529 2 $20,282 $16,694 -18
Total grant $17,028 $25,993 53
aid $15,432 $18,586 20
Need-based $14,684 $18,611 27
total grant $13,514 $13567 0.4
Institutional $13,187 $17,998 36
grant aid $11,960 $13,813 15
Total aid $20,110 $25,576 27 $22,446 $32,732 46
Number of 11 2
observations 7 8
White
Pricea $27,832 $35,099 26 $25,736 $30,952 20
Total grant $13,028 $12,512 -4
aid $6,711 $7,382 10
Need-based $11,645 $6,249 -46
total grant $5,271 $4,284 -19
Institutional $11,378 $10,187 -10
grant aid $5,105 $4,240 -17
Total aid $14,635 $14,130 -3 $20,042 $21,005 5
Number of 103 49
observations 81 48
Total 152 73
observations 112 72
Source: GAO analysis.
aPrice equals cost of attendance less total grant aid. Cost of attendance
equals tuition and fees, plus expenses (including room and board, and
books).
bData for other race, including Native American, unidentified race, and
foreign students were too few to report.
Notes: All values are (probability) weighted averages, and the monetary
values are in 2005 dollars.
Model Specifications and Estimation Methodology
Our econometric analysis is based on panel data, which pooled
cross-sectional and time series data. The cross-sectional data were based
on freshmen who enrolled in CA schools and non-CA schools, and the time
series data were for academic years 1995-1996 and 2003-2004. Where
feasible, we used panel-data estimation appropriate for cross-sectional
and time series data. Also, we used fixed-effects estimation instead of
random-effects estimation because the observations were not randomly
chosen and there were likely to be unobserved school-specific effects.29
The reported estimates were based on the fixed-effects estimators, using
probability weights, and the standard errors were robust.30
29The panel data were unbalanced because there were different observations
on the freshmen for each school in each academic year. An important
purpose in combining cross-sectional and time series data was to control
for individual school-specific unobservable effects, which may be
correlated with the covariates in the models. An advantage of using the
fixed-effects estimator was that there was no need to assume that the
unobserved school-specific effects were independent of the covariates.
However, unlike the random-effects estimator, the fixed-effects estimator
did not allow the inclusion of time-invariant variables, such as the
former Overlap group and membership in sports associations, as covariates.
30The weights are the probability weights from the number of students in
the sample for each school, and the robust estimates of the standard
errors are based on the Huber/White sandwich estimator. All estimates were
obtained using Stata.
Price, Tuition, and Financial Aid Equations:
Let Yijt be the dependent variable for freshman i's outcomes at the chosen
school j in academic year t, where the main outcome variable studied is
affordability represented by price (PRICEijt).31 The regression equations
were specified generally as follows:
(1)
where I and S are vectors of school (institution)-level and student-level
variables, and EMCA represents the consensus approach implementation; q
(time specific fixed-effects) and th (school specific fixed-effects) are
scalar parameters, and a and e are the constant and the random error
terms, respectively. There are interactions between EMCA and the
school-level variables and between EMCA and the student-level variables.32
We were primarily interested in the total effects of the implementation of
the consensus approach on affordability, as well as the effects that were
specific to particular groups of students, such as low-income and minority
students, and students who applied for financial aid.
Using equation 1, the total effect of the CA implementation on price was
estimated bywhere and are averages of I and S taken over the observations
for the CA schools during the period of the consensus approach
implementation (2003-2004).33 This measures the effect of the consensus
approach implementation on CA schools, relative to non-CA schools,
controlling for time invariant differences in schools and other variations
over time that are common to both groups. The coefficient measures the
unconditional effect of the consensus approach implementation on price,
while and measure the conditional effects of the consensus approach
implementation on price through the school-level variables and
student-level variables, respectively.
31The same model specification is used to estimate the financial aid
equations, and tuition equation, which excludes the student-level
variables.
32In the estimated equations, the interaction terms between EMCA and other
variables have the suffixes "*;" for example ENDOWSTU* is the interaction
term between ENDOWSTU and EMCA.
33This effect can be tested as a linear restriction if the joint test of
significance of EMCA and the terms involving EMCA is significant.
The expression for the total effects of the consensus approach
implementation can be evaluated for particular groups of students by
averaging I and S over that particular subset of students. For example,
the effects of the consensus approach implementation on prices paid by
low-income (INCLO) students can be estimated by where the school-level and
student-level variables are averaged over the low-income students. More
specifically, the second term is the coefficient estimates of each
school-level variable multiplied by the school-level variable averaged
over the subset of low-income (INCLO) students attending CA schools after
the consensus approach implementation; similarly the average is taken for
the third term, which is for the student-level variables.
Alternatively, we can use equation 1 to illustrate the effects of the
consensus approach implementation for particular groups. Consider a simple
example in which there are two student characteristics, and, where is an
indicator variable equal to one if the student is a financial aid
applicant and zero otherwise, andis an indicator equal to one if the
student is black, and zero otherwise. Then, using equation 1, the equation
for this example is:
(1.1)
Now consider a white student who is a financial aid applicant in school j
at time t.34 The predicted price for a white student if j is a CA school
is:
34White students are excluded from the race groups in the estimation to
avoid the dummy trap.
(1.2)
and the predicted price if j is not a CA school is:
(1.3)
The effect of the consensus approach implementation for a financial aid
applicant at school j is then the difference between equations 1.2 and
1.3, which is:
(1.4)
The coefficient measures the effect of adopting the consensus approach
that is invariant across school and student type, the term captures the
differential effect of adopting the consensus approach for a school with
characteristics Ijt, and the third term, , captures the differential
effect of adopting the consensus approach for a white student who is a
financial aid applicant. Repeating the exercise above for a black student
who is a financial aid applicant, the predicted effect of adopting the
consensus approach would be:
(1.5)
The first three terms in equation 1.5 are the same as equation 1.4, while
the fourth term captures the differential effect of the consensus approach
implementation for a black student. In this example, then, the estimated
effect of the consensus approach implementation on financial aid students
would be the weighted average of the terms in equations 1.4 or 1.5, with
weights corresponding to the proportions of white and black financial-aid
students across all schools j that adopted the consensus approach at time
t, respectively.
Another estimate of the consensus approach's effect on a particular group
is the estimated differential effect on a group, given by, holding
everything else constant. For example, one can ask how a low-income
student as compared to a high-income student would be affected by the
consensus approach implementation, assuming all other characteristics of
the student and the student's school are held constant. This estimated
effect is simply given by the element of the vector in that corresponds to
INCLO. This differs from the total effect of the consensus approach
implementation discussed above by taking as given the consensus approach
implementation, and by abstracting from the likelihood that low-income
students will have other characteristics and attend different schools than
non low-income students. We will also discuss the coefficient, which
captures the value of the dependent variable for the particular group in
both CA and non-CA schools before the consensus approach implementation,
where necessary.
The total effect of the exemption on price as well as its specific effects
on particular groups will depend on which theory of the exemption is
supported by the data. In particular, we expect price to be lower for
disadvantaged students if the social benefit theory is valid; on the other
hand, price will increase if the anti-competitive theory is valid.
Similarly, the effects of the student-level variables would depend on the
theories of the effects of the exemption. For the effects of the
school-level variables, ENDOWSTU should be negative because with more
resources there is less need to raise tuition and there will be more funds
for grant aid. RANKAVG should be negative because as the quality of the
school decreases tuition as well as grant aid should decrease. ENROLUG
would be negative if higher growth in student enrollment perhaps means
more revenues and less need to raise tuition. On the other hand, if
students' education is on net subsidized by other sources of school income
then ENROLUG would be positive as increased enrollment increases the costs
to the school of providing education. And TENURED should be positive if
more tenured faculty implies higher quality.35
We estimated equation 1 for price, as well as for tuition and the
financial aid variables, using probability-weighted regression and robust
standard errors, as well as the fixed-effects estimator for panel data.36
See the regression estimates for price and tuition in table 14, and those
for the financial aid variables in table 15.37
35The effects of these variables on tuition were expected to be similar to
that of price. On the other hand, the effects of these school-level
variables on the financial aid variables were expected to be opposite to
that of price.
36The statistical procedure we used is AREG in Stata.
The regression models for the price, tuition, and financial aid variables
are all highly significant using the F-values of the models. See tables 14
and 15. Furthermore, the school-level variables generally have the
expected effects. In particular, for the price equation, a student
enrolled in a school with an endowment per student (ENDOWSTU) of $250,000
paid about $5,000 lower price.38 Also, a student paid about $464 less for
a school with a unit drop in its selectivity (RANKAVG). Although the
effect is not significant, the positive sign for ENROLUG suggests that an
increase in enrollment growth may result in a higher price paid, implying
that education is net subsidized and increases in enrollment increases the
cost of providing education; and vice versa. Finally, a student enrolled
in a school with 10 percent higher tenured faculty (TENURED) paid about
$3,310 higher. As discussed earlier, the effects of the student-level
variables depend on which theory of the effects of the higher education
exemption is relevant.39
Student enrollment equation:
The regression equation for enrollment into a CA school (ENRCAijt) would
depend on student characteristics. Generally, enrollment is the outcome of
decision-making that included application, admission, and acceptance of
the admission offer. The first and third decisions are made by the
student, and the second decision is made by the school. Therefore, in
general, both student-level variables and school-level variables would be
relevant. However, our approach, as indicated in equation 2, treated the
CA schools essentially the same and likewise for the non-CA schools, with
differences between the two groups other than the consensus approach
implementation captured by the constant term in the regression. The
enrollment equation was thus specified as follows, excluding school-level
variables as regressors:
37The regression estimates for the financial aid variables excluded non
financial-aid applicants, which reduced the number of observations but not
the number of schools. Similar results were obtained for the price
equation when the estimates were based on only financial-aid applicants.
The regression estimates for tuition were obtained by excluding
student-level variables because students at a school were charged the same
tuition.
38The $4,800 decrease is approximately equal to $250,000 x -(0.01935).
39As discussed earlier, similar arguments can be obtained for the tuition
and financial aid variables.
(2)
F is the standard normal cumulative probability distribution function.
Similar to equation 1, equation 2 includes student characteristics (with
coefficients r), time fixed-effects captured by AY2003, and the
interaction of the time variable AY2003 with student characteristics (with
coefficients g).40
The time specific fixed-effect for AY2003 captures any shift, which is
constant across students, toward or away from the CA schools, after the
consensus approach implementation, while the interaction terms between the
AY2003 and the student characteristics capture shifts toward or away from
the CA schools by students with specific characteristics.41
The marginal effect of the consensus approach implementation is captured
by the effects of AY2003 on enrollment in CA schools. Specifically, this
equals,
where f is the standard normal probability density function. It should be
noted that if AY2003 affects the probability of enrollment in CA schools,
it would be a valuable suggestive evidence about the potential impact of
the consensus approach implementation. However, it would not establish
that the consensus approach implementation caused the shift. This is
because it is possible that such effects might be due to changes in other
factors at CA schools versus non-CA schools (e.g., more rapid endowment
growth in the latter than the former). The effect of the consensus
approach implementation is the change in the probability of enrollment in
CA schools relative to non-CA schools as a result of the consensus
approach implementation. The overall effect of the CA implementation as
well as the effects of the consensus approach implementation on particular
groups of students, such as low-income students and those who applied for
financial aid, can be obtained similar to the discussion above for the
price.
40The model could not be estimated with school specific fixed-effects
because they predict successes or failures perfectly.
41In the estimated equations, the interaction terms between AY2003 and
other variables have the suffixes "*;" for example INCLO* is the
interaction term between INCLO and AY2003.
The marginal effect of the student characteristics is captured by the
effects of S on enrollment in CA schools. Specifically, this equals,
The effect of the consensus approach implementation on how the
probabilities of enrollment of low-income and minority students, and those
who applied for financial aid, are affected can be obtained similar to the
discussion for the price.
Similar to the discussion for the price equation, the effects of the
exemption and the student-level variables on enrollment into CA schools
will depend on which theory of the exemption is valid. In particular, the
social benefit theory will imply increased likelihood of enrollment into
CA schools, especially of low-income students, because prices will be
lower. While the opposite will occur with the anti-competitive theory
because average price will be higher.
We estimated equation 2 for student enrollment using the probit
estimation, with probability weights and robust standard errors.42 The
regression estimates are in table 14.
The regression model for enrollment in table 14 is significant using the
chi-square of the model. As indicated earlier, we expect the estimation
results will enable us to determine if the likelihood of enrollment into
schools implementing the consensus approach by various student groups is
more consistent with the social benefit theory or the anti-competitive
theory of the effects of the higher education exemption.
42We could not use the panel data estimation technique for probit
(XTPROBIT) because of lack of convergence. Similar results were obtained
when the estimates were based on only students who applied for
financial-aid.
Table 14: Regression Estimates of Effects of Consensus Approach
Implementation on Price, Tuition, and Enrollment
Variable Price Tuition Enrollment
EMCA 43,743.75a -1,720.72 [0.512] n/a
[0.002]
AY2003 12,133.55a 5,650.28a [0.000] 0.25751 [0.631]
[0.000]
Student-level
FINAID -7,573.33a n/a 0.01292 [0.854]
[0.000]
FINAID* 190.21 [0.911] n/a -0.30976 b
[0.015]
ASIAN 1,084.13 n/a -0.01069
[0.363] [0.910]
ASIAN* -7,109.66c n/a -0.08486
[0.055] [0.613]
BLACK -9,444.68a n/a 0.19398 [0.129]
[0.000]
BLACK* -566.20 [0.921] n/a -0.20422
[0.339]
HISPANIC -4,686.31a n/a 0.06790 [0.574]
[0.007]
HISPANIC* -1,610.12 n/a 0.35354b
[0.588] [0.021]
FOREIGN 3,398.09 n/a -0.25968
[0.230] [0.221]
FOREIGN* 10,588.17b n/a 0.36734c
[0.026] [0.054]
OTHER -2,047.09 n/a 0.02366 [0.887]
[0.407]
OTHER* -1,077.16 n/a 0.30781c
[0.727] [0.059]
EFC 0.10871a n/a 2.30e-06
[0.000] [0.190]
EFC* -0.00745 n/a -3.43e-06
[0.872] [0.219]
INCLO -8,427.06a n/a -0.15253
[0.000] [0.196]
INCLO* -6,507.64 n/a -0.15185
[0.274] [0.489]
INCLOMD -8,696.19a n/a -0.03045
[0.000] [0.797]
INCLOMD* -1,789.68 n/a 0.16593 [0.384]
[0.593]
INCMD -4,804.25a n/a 0.06859 [0.506]
[0.000]
INCMD* 945.64 [0.771] n/a -0.16166
[0.303]
INCUPMD -1,434.39 n/a -0.03378
[0.163] [0.689]
INCUPMD* -1,715.73 n/a -0.07025
[0.522] [0.657]
SCORESAT -2.48 [0.430] n/a 0.00024 [0.227]
SCORESAT* -7.82 [0.383] n/a 0.00013 [0.764]
School-level
ENDOWSTU -0.01935a -0.00401a[0.008] n/a
[0.001]
ENDOWSTU* -0.01056b 0.00038 [0.831] n/a
[0.044]
RANKAVG -464.33c -151.06c [0.064] n/a
[0.051]
RANKAVG* -798.05a 36.28 [0.792] n/a
[0.000]
ENROLUG 19,038.16 35,553.57 [0.133] n/a
[0.785]
ENROLUG* -45,843.9 -44,159.24c [0.095] n/a
[0.518]
TENURED 33,100.97a -610.06 [0.492] n/a
[0.000]
TENURED* -25,823.9b 4,632.48 [0.315] n/a
[0.014]
Constant 29,651.89a 28,746.01a [0.000] n/a
[0.000]
Test statistic of 22.97a [0.000] 375.71a [0.000] 37.68b [0.050]
modeld
R-squared 0.62 0.99 0.06
Sample size 518 28 518
Joint test for EMCA 2.85a [0.000] 1.15 [0.460] 18.70b [0.133]
Linear restriction 1.18 [0.240] -0.59 [0.586] n/a
test for EMCAe
Source: GAO analysis.
aStatistically significant at the 1 percent level or lower. P-values are
in brackets.
bStatistically significant at the 5 percent level or lower. P-values are
in brackets.
cStatistically significant at the 10 percent level or lower. P-values are
in brackets.
dF-statistic values for the price and tuition equations, and chi-square
values for the enrollment equation.
et-statistic values for the price and tuition equations, and z-statistic
values for the enrollment equation.
Notes: N/A means data are not available or applicable.
* means interaction terms with EMCA for price and tuition equations, and
interaction terms with AY2003 for the enrollment equation.
Estimates of price and tuition are obtained using fixed-effects models.
Estimates for enrollment are the marginal effects from a probit model.
Table 15: Regression Estimates of Effects of Consensus Approach
Implementation on Financial Aid
Need-based grant
Variable Total grant aid aid Total aid
EMCA -28,705.63c -15,512.93 -50,805.28b
[0.066] [0.178] [0.016]
AY2003 2,159.46 [0.348] -1,511.37 [0.522] 7,385.03c
[0.062]
Student-level
FINAID n/a n/a n/a
FINAID* n/a n/a n/a
ASIAN -722.86 [0.649] -740.76 [0.590] -2,638.73
[0.221]
ASIAN* 6,970.89 [0.142] 6,546.91c [0.066] 8,275.76 [0.221]
BLACK 7,914.65a [0.000] 4,796.63b [0.026] 8,425.58a
[0.000]
BLACK* 2,546.02 [0.656] -2,267.21 [0.514] 965.32 [0.862]
HISPANIC 4,709.93b [0.023] 2,826.32 [0.140] 2,281.05 [0.389]
HISPANIC* 2,038.96 [0.570] 206.79 [0.947] 2,606.82 [0.622]
FOREIGN -5,754.55b -4,634.40 [0.132] -11.961.86a
[0.031] [0.002]
FOREIGN* -10,965.98b -10,414.55b -11,659.21c
[0.032] [0.032] [0.068]
OTHER 4,656.42 [0.169] 5,229.39 [0.129] 4,196.89 [0.349]
OTHER* -2,813.57 [0.481] -1,725.88 [0.667] -4,098.47
[0.496]
EFC -0.16058a [0.000] -0.149889a -0.187923a
[0.000] [0.000]
EFC* 0.044382 [0.397] 0.03185 [0.472] 0.006624 [0.929]
INCLO 7,178.94a [0.000] 7,932.99a [0.000] 3,276.04 [0.177]
INCLO* 7,346.13 [0.239] 6,955.88c [0.096] 14,970.89
[0.020]
INCLOMD 8,227.84a [0.000] 8,747.21a [0.000] 7,153.01a
[0.001]
INCLOMD* 2,830.62 [0.395] 3,585.08 [0.198] 3,678.72 [0.359]
INCMD 5,084.33a [0.000] 3,488.27b [0.015] 4,116.59b
[0.040]
INCMD* 1,835.35 [0.617] 4,773.69 [0.134] 5,366.73 [0.339]
INCUPMD 2,215.09c [0.093] 1,699.48 [0.161] 1,643.62 [0.419]
INCUPMD* 600.56 [0.851] -473.12 [0.873] 2,503.61 [0.632]
SCORESAT 1.23 [0.729] -0.51436 [0.882] -3.05 [0.561]
SCORESAT* 10.81 [0.306] 1.97 [0.775] 18.18 [0.170]
School-level
ENDOWSTU 0.00684 [0.317] 0.01056 [0.144] -0.00583 [0.512]
ENDOWSTU* 0.001786 [0.774] 0.00272 [0.644] 0.015595c
[0.092]
RANKAVG 88.79 [0.773] 406.70 [0.114] 58.30 [0.888]
RANKAVG* 140.56 [0.571] 38.06 [0.859] 566.66 [0.178]
ENROLUG -36,447.82 -238,689.4a -12,713.12
[0.623] [0.004] [0.922]
ENROLUG* 30,780.66 [0.683] 246,969.5a 32,081.77
[0.003] [0.807]
TENURED -4,340.32 [0.545] 5,070.34 [0.486] 8,685.55 [0.530]
TENURED* 15,405.02 [0.204] 16,399.34 [0.103] 11,726.97
[0.514]
Constant 8,727.25 [0.375] -1331.71 [0.878] 17,640.66
[0.209]
F value of model 14.81a [0.000] 16.36a [0.000] 8.79a [0.000]
R-squared 0.55 0.54 0.36
Sample size 409 409 409
F value of joint test 1.13 [0.328] 1.83b [0.026] 1.60c [0.067]
t value of linear n/a 2.05b [0.041] 0.64 [0.525]
restriction
Source: GAO analysis.
aStatistically significant at the 1 percent level or lower. P-values are
in brackets.
bStatistically significant at the 5 percent level or lower. P-values are
in brackets.
cStatistically significant at the 10 percent level or lower. P-values are
in brackets.
Notes: n/a means data are not available or applicable.
*Means interaction terms with EMCA.
Estimation Results of the Effects of Attending Meetings and Implementing the
Consensus Approach43
The results of estimating equations 1 and 2 for the total effects of the
CA implementation on affordability and enrollment are summarized in table
16, based on the regression results in tables 14 and 15. The results for
price and enrollment in table 16 contain the key findings of the entire
study, with the other variables (tuition and financial aid) providing
information that supplements the findings for price. 44
43Due to lack of sufficient data, we could not obtain separate estimates
of the effects of attending meetings only or the effects of implementing
the consensus approach only because it involved only two schools-Brown and
Stanford. Also, we could not obtain separate estimates of the effects of
implementing fully or partly the consensus approach because only three of
the seven CA schools in sample 2 (Georgetown, Vanderbilt, and Wake Forest)
had not fully implemented the CA.
44All tests are performed using the 5 percent or lower level of
significance.
Total Effects of Implementing the Consensus Approach (from table 16)
Prices:45
For the average student, the consensus approach implementation did not
significantly change the prices paid by students in CA schools compared to
non-CA schools, including the effects on low-income and minority students
and students who applied for financial aid.46
Tuition:47
The CA schools, compared to non-CA schools, did not significantly change
the tuition they charged students as a result of the consensus approach
implementation.
Total grant aid:48
The consensus approach implementation did not significantly change the
amount of total grant aid received by students in CA schools compared to
non-CA schools.
Need-based total grant aid:49
45The results were similar when we limited the data to only students who
applied for financial aid.
46The effect of the consensus approach implementation on lower-middle
income was positive and significant at the 10 percent level. We performed
several tests for the total effects of the consensus approach on prices.
First, the effect was significant at the 5 percent level when data for
only students who applied for financial aid were used. Second, the total
effect of the CA on prices was $3,488 and significant at the 5 percent
level when ENROLUG and ENROLUG* were excluded from the model. Third,
because prices are bounded at the lower end at zero and at the upper end
at the cost of attendance, we also estimated the price equation using
Tobit regressions. The total effect of the consensus approach on prices
was negative and insignificant (at the 10 percent level). Unlike the
fixed-effects estimates, the Tobit estimates were unweighted and the
standard errors were not robust.
47The results are based on the seven CA and the seven non-CA schools in
tables 11 and 12. Similar results were obtained when we included the
schools that had no SAT scores in AY 2003-2004-three CA schools (Boston,
MIT, and Pennsylvania) and two non-CA schools (Tufts and Yeshiva).
48The value of the effect of the CA on institutional grant aid was $1,331,
but not significant.
49The effect of the CA on need-based institutional aid was generally
similar to need-based total grant aid. The effect was about $6,020 and
significant at the 5 percent level, with a confidence interval of between
$512 and $11,528.
The consensus approach implementation increased the amount of need-based
total grant aid received by students in CA schools compared to non-CA
schools by about $6,125, with a confidence interval of between $239 and
$12,011.50 The amounts of need-based grant aid received by students in CA
schools compared to non-CA schools were higher for middle income students
by about $20,221, with a confidence interval of between $6,718 and
$33,724. Asian students received higher need-based grant aid of about
$14,628, with a confidence interval of between $5,051 and $24,206;
Hispanic students received higher need-based grant aid of about $9,532,
with a confidence interval of between $1,006 and $18,059; and white
students received higher need-based grant aid of about $6,017, with a
confidence interval of between $178 and $11,856.
Total aid:51
The consensus approach implementation did not significantly change the
amount of total aid received by students in CA schools compared to non-CA
schools. However, low-income students in CA schools received higher total
aid of about $12,121, with a confidence interval of between $1,837 and
$22,404.52
Enrollment:
The consensus approach implementation did not significantly change the
overall likelihood of enrollment into CA schools compared to non-CA
schools, for all types of students.
50The value of the effect of the CA on non-need-based grant aid was
estimated to be about -$6,873, though not significant; the F-test of the
joint significance of EMCA and its interactive terms had a p-value of 14
percent, and the test of the total effect of the CA had a p-value of 2.1
percent.
51The value of the effect of the CA implementation on self-help aid
(loans, including PLUS, and work study) was $1,034, but not significant.
52The value of the total effect of the CA on total aid was estimated to be
about $7,140, though not significant; the F-test of the joint significance
of EMCA and its interactive terms had a p-value of 20 percent, and the
test of the total effect of the CA had a p-value of 1.4 percent.
Table 16: Estimates of Effects of Consensus Approach Implementation on
Affordability and Enrollment in CA Schools Relative to Non-CA Schools
Total effect
of consensus Need-based Probability
approach total of
on... Price Tuition Total grant aid grant aid Total aid enrollment
All students $3,021 -$433 -$749 $6,125b -$2,886 38%
[-$2,026, [-$2,465, [-$6,967, [$239, [-$11,805, [8%, 67%]
$8,068] $1,599] $5,470] $12,011] $6,034]
Financial-aid $2,177 n/a n/a n/a n/a 22%
applicants
[-$3,319, [-11%,
$7,673]
54%]
Low income -$4,061 n/a $3,688 $1,956 $12,121b [-52%,
[-$15,583, [-$8,511, [-$5,232, [$1,837, 170%]
$7,461]
$15,887] $9,144] $22,404]
Lower-middle $8,089c n/a -$3,671 $6,556 -$7,776 95%
income
[-$263, Higher Education [-$2,145, [-$19,776, [6%,
$16,441] [-$12,487,Higher
Education Higher $15,257] $4,224] 184%]
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education Higher
Education
$5,145]
Middle income $2,320d n/a $1,618 $20,221a $1,178 26%
[-$8,043, [-$11,221, [$6,718, [-$19,616, [-41%,
$12,682]
$14,457] $33,724] $21,971] 93%]
Upper-middle -$1,048 n/a -$973 $2,769 -$3,054 18%
income
[-$7,641, [-$7,801, [-$3,986, [-$13,177, [-47%,
$5,545]
$5,855] $9,524] $7,068] 82%]
High income $3,699 n/a -$714 $4,687c -$3,856 31%
[-$824, [-$6,905, [-$449, [-$12,817, [-6%,
$8,222]
$5,476] $9,824] $5,104] 68%]
Asian -$376 n/a $5,726 $14,628a $3,694 1%
students
[-$10,426, [-$5,671, [$5,051, [-$13,693, [-78%,
$9,674]
$17,123] $24,206] $21,082] 80%]
Black $4,468 n/a -$1,227 $4,332 -$6,542 -26%
students
[-$7,452, [-$13,238, [-$4,992, [-$20,353, [-142%,
$16,387]
$10,783] $13,657] $7,269] 91%]
Hispanic $1,168d n/a $1,520 $9,532b $3,648 108%
students
[-$6,744, [-$8,300, [$1,006, [-$8,981, [-6%,
$9,079]
$11,341] $18,059] $16,278] 222%]
White $2,588 n/a -$491 $6,017b -$2,879 19%
students
[-$2,403, [-$6,766, [$178, [-$11,922, [-14%,
$7,578]
$5,784] $11,856] $6,164] 52%]
Number of 518 28 409 409 409 518
observations
Schools
CA Schools: Cornell University, Duke University, Georgetown University, University of
Notre Dame, Vanderbilt University, Wake Forest University, Yale University
Non-CA Schools: Brandeis University, Bryn Mawr College, New York University,
Princeton University, Tulane University, University of Rochester, Washington
University at St. Louis
Source: GAO analysis.
aStatistically significant at the 1 percent level or lower.
bStatistically significant at the 5 percent level or lower.
cStatistically significant at the 10 percent level or lower.
d Effects were negative when data for only financial aid applicants were
used.
Notes: The values in brackets are the 95 percent confidence intervals for
the estimates that are significant at the 5 percent level or lower.
"n/a" means data are not available or applicable.
All the monetary values are in 2005 dollars.
Results are based on tables 14 and 15.
The calculated values are based on where the average values are for all
students.
The estimates are based on where the average values are for the relevant k
subgroup of students.
Prior Levels and Differential Effects of the Consensus Approach on Affordability
and Enrollment for Students with Particular Characteristics53
We discuss the estimates of affordability and the likelihood of enrollment
in both the schools that adopted the consensus approach and those that did
not, of students with particular characteristics, before the consensus
approach was implemented. The estimates are reported in table 17, based on
tables 14 and 15. These estimates could help explain the extent to which
the consensus approach affected particular groups of students. For
instance, if certain students were receiving higher financial aid awards
prior to the consensus approach, they may be less likely to receive much
higher awards as a result of its adoption. We also discuss the
differential effects on students with particular characteristics that the
consensus approach may have had on affordability and enrollment at those
schools. The estimates are reported in table 18, based on tables 14 and
15. As already discussed, these estimates indicate how the consensus
approach affected students with particular characteristics, assuming all
the other characteristics of the students are held constant.
53The results for financial aid applicants are relative to non financial
aid applicants, those for the income groups are relative to the
high-income students, and those for the racial groups are relative to the
white students.
Prices:
Some students paid lower prices prior to the CA implementation; in
particular, financial aid applicants relative to non-financial aid
applicants; low income, lower-middle income, middle-income students
relative to high-income students; and black and Hispanic students relative
to white students. But there were no significant differential effects of
implementing the consensus approach on prices paid by these groups of
students in CA schools.
Total grant aid:
Some students received higher total grant aid prior to the consensus
approach implementation; in particular, low-income, lower-middle income,
middle-income, black, and Hispanic students.
Need-based total grant aid:
Some students received higher need-based aid prior to the consensus
approach implementation; in particular, low-income, lower-middle income,
middle-income, and black students. But there were no significant
differential effects of implementing the consensus approach on prices paid
by these groups of students.
Total aid:
Some students received higher total aid prior to the consensus approach
implementation; in particular, middle-income, and black students. But
lower-middle income students received lower total aid prior to the
consensus approach implementation. Only low-income students in CA schools
received higher aid, compared to high-income students, as a result of
implementing the consensus approach.
Enrollment:
Students generally were not more or less likely to enroll in a CA school
prior to the consensus approach implementation. However, implementing the
consensus approach lowered the likelihood of enrollment of financial-aid
students, compared to non-financial aid applicants, while the likelihood
of enrollment of Hispanic students increased, compared to white students,
in CA schools.
Table 17: Estimates of Affordability and Enrollment before the Consensus
Approach Implementation for Particular Groups of Students in Both CA and
Non-CA Schools
Need-based
Total total grant Probability
Students Price grant aid aid Total aid of enrollment
Financial-aid
applicantsd -$7,573a N/A N/A N/A 1%
Low incomee -$8,427a $7,179a $7,933a $3,276 -15%
Lower-middle
incomee -$8,696a $8,228a $8,747a -$7,153a -3%
Middle incomee -$4,804a $5,084a $3,488b $4,117b 7%
Upper-middle
incomee -$1,434 $2,215c $1,699 $1,644 -3%
High income N/A N/A N/A N/A N/A
Asian studentsf $1,084 -$723 -$740 -$2,639 -1%
Black studentsf -$9,445a $7,915a $4,797b $8,426a 19%
Hispanic
studentsf -$4,686a $4,710b $2,826 $2,281 7%
White students N/A N/A N/A N/A N/A
Number of
observations 518 409 409 409 518
Schools CA Schools: Cornell University, Duke University,
Georgetown University, University of Notre Dame,
Vanderbilt University, Wake Forest University, Yale
University
Non-CA Schools: Brandeis University, Bryn Mawr College,
New York University, Princeton University, Tulane
University, University of Rochester, Washington University
at St. Louis
Source: GAO analysis.
aStatistically significant at the 1 percent level or lower.
bStatistically significant at the 5 percent level or lower.
cStatistically significant at the 10 percent level or lower.
dThe estimates are relative to non-financial aid applicants.
eThe estimates are relative to high income students.
fThe estimates are relative to white students.
Notes: Results are from tables 14 and 15, based on the coefficient in
equations 1 and 2. For instance, the value for price for financial-aid
applicants is based on the estimated coefficient FINAID in table 14.
N/A means data are not available or applicable.
All the monetary values are in 2005 dollars.
Table 18: Differential Effects of Consensus Approach Implementation on
Affordability and Enrollment in CA Schools for Particular Groups of
Students
Need-based
Total total grant Probability
Students Price grant aid aid Total aid of enrollment
Financial-aid
applicantsd $190 N/A N/A N/A -31%b
Low-incomee -$6,508 $7,346 $6,956c $14,971b -15%
Lower-middle
incomee -$1,790 $2,831 $3,585 $3,679 17%
Middle incomee $946 $1,835 $4,774 $5,367 -16%
Upper-middle
incomee -$1,716 $601 -$473 $2,504 -7%
High income n/a n/a n/a n/a n/a
Asian studentsf -$7,110c $6,971 $6,547c $8,276 -9%
Black studentsf -$566 $2,546 -$2,267 $965 -20%
Hispanic
studentsf -$1,610 $2,039 $207 $2,607 35%b
White students n/a n/a n/a n/a n/a
Number of
observations 518 409 409 409 518
Schools CA Schools: Cornell University, Duke University,
Georgetown University, University of Notre Dame,
Vanderbilt University, Wake Forest University, Yale
University
Non-CA Schools: Brandeis University, Bryn Mawr College,
New York University, Princeton University, Tulane
University, University of Rochester, Washington University
at St. Louis
Source: GAO analysis.
aStatistically significant at the 1 percent level or lower.
bStatistically significant at the 5 percent level or lower.
cStatistically significant at the 10 percent level or lower.
dThe estimates are relative to non-financial aid applicants.
eThe estimates are relative to high income students.
fThe estimates are relative to white students.
Notes: Results are from tables 14 and 15, based on the coefficient in
equations 1 and 2. For instance, the value for price for financial-aid
applicants is based on the estimated coefficient FINAID* in table 14.
N/A means data are not available or applicable.
All the monetary values are in 2005 dollars.
Limitations of the Study
Sample Selection bias
The findings of the study could be limited by the potential of selection
bias if the CA schools had characteristics that we could not control for
that made them more inclined to adopt the consensus approach and
independently influenced the outcome variables. We believe that this is
not a serious problem with the estimation since the
difference-in-difference approach includes CA schools before the
implementation of the CA, implying the latter selection problem would
require significant change over a short time span in the character of
these schools. Furthermore, a key factor that might motivate schools to
join the 568 Group is the legacy of the Overlap group. The 568 Group has
objectives that are similar to those stated by the Overlap group-to be
able to offer financial aid to more needy students. Our test indicated
that the chances of a former Overlap group member joining or not joining
the 568 Group did not differ between the two groups of schools in our
sample.54 Thus, the similarity between the two groups, in terms of a
school joining the 568 Group, implied the potential for selection bias may
be small.
Measures of Price
In our analysis, the total grant aid does not include self-help aid (loans
and work study). However, if the true amount of total grant aid should
include some proportion of self-help aid, then its exclusion would lead to
an underestimation of total grant aid. Nonetheless, we believe this did
not significantly affect our results since we found that the consensus
approach implementation did not affect self-help aid.
Early Decision Admissions
It may be that early admit students pay higher prices because early
decision admission might be used by need-blind schools as a screening
mechanism to indirectly identify a student's willingness-to-pay. Under the
early decision process a non-financial aid student is therefore more
likely to be admitted than a financial-aid student of comparable
quality.55 We did not expect the early decision process to affect our
results because while the process might help identify a student with a
higher willingness to pay, it is the student's ability to pay that
determines the need-based aid offered by the 568 Group. Furthermore, the
total probability of enrollment of a financial-aid applicant was similar
to that of a non-financial aid applicant both before and after the
consensus approach implementation, even though the consensus approach
implementation tended to decrease the likelihood of enrollment of
financial-aid students.
54We tested for the equality of the proportions of CA schools and non-CA
schools that were members of the former of the Overlap group. We used the
11 CA schools and the 14 non-CA schools in samples 1 through 4 in table 9.
The CA schools had 5 Overlap members and the non-CA schools had 3 Overlap
members.
55See Kim (2005).
Excluded Schools of Comparable Selectivity
We could not include all the schools affiliated with the 568 Group in the
analysis because of data limitations. (See the list of unmatched treatment
schools in table 9.) However, there were several similarities (in terms of
"best college" ranking, endowment, tuition and fees, and percentage of
tenured faculty) as well as differences (in terms of freshmen enrollment)
between the included and excluded CA colleges.
Limited Data Availability
The data were available for only one academic year period after
implementation of the consensus approach. This could mask potential
effects of the consensus approach since these effects could be gradual,
rather than immediate, and therefore take time to for the effects to be
captured. Also, the small sample size of the data could make the estimates
less precise, especially for some of the subgroups of students we
considered. However, we checked to ensure that the estimates were
consistent with the data by estimating the predicted values corresponding
to the observed mean values for price, the key variable of interest, and
the financial aid variables. The results, presented in table 19, show that
the predictions of our model are consistent qualitatively with the
observed data.
Table 19: Comparison of Observed and Predicted Price and Financial Aid
Variables in CA and Non-CA Schools: Pre- and Post-Consensus Approach
Implementation Period
CA Schools Non-CA Schools
1995-1996 2003-2004 Difference 1995-1996 2003-2004 Difference
Price
All students
Observed $28,039 $35,488 $7,449 $28,068 $30,838 $2,770
Predicted $30,791 $37,171 $6,380 $25,386 $27,882 $2,496
Financial-aid
Observed $25,845 $32,897 $7,052 $24,960 $29,705 $4,745
Predicted $28,222 $34,352 $6,130 $22,347 $27,330 $4,983
Non financial-aid
Observed $34,645 $44,504 $9,859 $37,714 $44,292 $6,578
Predicted $38,771 $46,625 $7,854 $35,127 $34,973 -$154
Low income
Observed $12,566 $10,095 -$2,471 $18,950 $21,886 $2,936
Predicted $14,272 $11,389 -$2,833 $13,613 $18,106 $4,493
Lower-middle-income
Observed $19,220 $30,437 $11,217 $17,623 $20,546 $2,923
Predicted $20,650 $32,075 $11,425 $15,156 $19,501 $4,345
Middle-income
Observed $24,785 $34,201 $9,416 $22,560 $26,069 $3,509
Predicted $26,035 $36,438 $11,838 $21,092 $20,764 -$328
Upper-middle-income
Observed $26,285 $32,310 $6,025 $29,429 $34,305 $4,876
Predicted $31,423 $35,121 $3,698 $26,566 $27,616 $1,050
High income
Observed $32,616 $39,496 $6,880 $33,137 $34,138 $1,001
Predicted $35,538 $41,043 $5,505 $31,129 $32,582 $1,453
Financial aid-
All students
Total grant aid
Observed $9,142 $9,775 $633 $13,391 $13,960 $569
Predicted $10,285 $11,877 $1,592 $12,181 $13,194 $1,013
Need-based grant
aid
Observed $7,771 $6,439 -$1,332 $11,863 $8,122 -$3,741
Predicted $7,443 $6,170 -$1,273 $12,277 $9,151 -$3,126
Total aid
Observed $16,604 $16,046 -$558 $19,827 $22,255 $2,428
Predicted $17,998 $17,957 -$41 $18,425 $20,127 $1,702
Source: GAO analysis.
Appendix III: Classification of
1999-2000 Academic Year and Schools Only Attending the 568 Group Meetings
We conducted tests to determine whether to use data collected in academic
year 1999-2000 and whether schools that attended meetings of the 568
President's group but did not implement the consensus approach could be
included in our analysis. First, the academic year 1999-2000 was very
close to the establishment of the 568 President's Group, which occurred in
1998. The 1999-2000 academic year might have been a transitional period,
and it would therefore not be appropriate to use the data as part of the
period before the 568 Group implemented the consensus approach. Second,
there were five schools, among the schools with data available for our
econometric analysis, that either only attended the 568 Group meetings
(Case Western Reserve University, Stanford University, and University of
Southern California) or were members of the 568 Group but had not
implemented the CA as of 2003 (Brown University and Dartmouth College). We
therefore investigated which group-control or treatment-each of the five
schools belonged.
Does Academic Year 1999-2000 belong to the Pre- or Post- Consensus Approach
Implementation Period?
We used the data for sample 4 to investigate if data collected in
1999-2000 belonged in the pre-CA period (with data collected in
1995-1996). Although both samples 1 and 4 have data for 1995-1996 and
1999-2000, we chose sample 4 because it was the larger sample. See table 9
in appendix II for the list of the schools in each sample and the academic
years for which data were available.
The tests were performed using the Chow test, which is of the form:1
(1) y = b01 + b11 x1 + b21 x2 + u, u ~ N(0,s2), for group = 1995-1996
(g1), and
(2) y = b02 + b12 x1 + b22 x2 + u, u ~ N(0,s2) for group = 1999-2000 (g2).
Pooling the two groups of data we estimated,
(3) y = b01 + b11 x1 + b21 x2 + (b02-b01)g2 + (b12-b11)g2x1 +
(b22-b21)g2x2 + u, where g2 is an indicator variable.
The test examines the hypothesis that the added coefficients are jointly
zero: (b02-b01) = (b12-b11) = (b22-b21) = 0.
An insignificant test statistic (a small test statistic and a large
p-value) suggests that the above equality holds, and there is no
difference between the estimates for 1999-2000 and the group with which it
is compared (1995-1996). On the other hand, a significant statistic (a
large test statistic and a small p-value) suggests that the above equality
does not hold and the 1999-2000 is different from the group with which it
is compared (1995-1996).
1See http://www.stata.com/support/faqs/stat/chow3.html for details.
We combined 1999-2000 with 1995-1996 and tested if the coefficients for
1999-2000 differed from that of 1995-1996, using sample 4. The tests were
done for price, the key variable affecting student outcomes for schools.
We performed a joint test that the added coefficients in equation 3 are
jointly zero. The F-value is 1.71, and significant with a p-value of
0.0375. This implied that data collected in 1999-2000 did not belong to
with the 1995-1996 data in the pre-CA period.2
Similarly, we examined if 1999-2000 belonged to the post-CA period by
combining 1999-2000 with 2003-2004, using sample 3. The F-value of the
joint test is 8.36, and significant with a p-value of 0.0. This implied
that 1999-2000 data did not belong to with the 2003-2004 data in the
post-CA period.
These results suggest that it was more appropriate to exclude 1999-2000
from the analysis, implying that samples 1 and 2, which have data for the
pre-CA period (1995-1996) and the post-CA period (2003-2004) would be more
appropriate. However, because sample 2 was larger than sample 1, our
subsequent analysis used sample 2.
Do the Schools That Only Attended the 568 group Meetings belong to the Control
or Treatment Group?
We performed an analysis similar to that described above to determine
whether schools that only attended meetings-Brown University, Case Western
Reserve University, Dartmouth College, Stanford University, and University
Southern California (USC)-belonged in the treatment or control group. We
determined whether the behavior of each of these schools was more
consistent with the control schools or the treatment schools after the
consensus approach implementation, using data for 2003-2004. Since we had
determined from the above analysis that samples 1 and 2 are more
appropriate for our subsequent analysis, we focus on sample 2, the larger
sample, for these tests.3
2As expected, the estimates from the pooling (equation 3) are the same as
for the separate estimates (equations 1 and 2). Also the residual
variances from equations 1 and 2 were similar, suggesting that the pooling
was appropriate. This applies to all the other Chow tests we performed.
To Which Group Did Brown Belong-Control or Treatment?
Similar to the analysis in section above, we included Brown in the control
group and tested if the coefficients for Brown differed from the control
group. We performed a joint test and obtained an F-value of 25.68,
significant at 0.00. This implied that Brown did not belong to the control
group. For the treatment group test, the F-value was 7.37, significant at
0.00. This also implied that Brown did not belong to the treatment group.
Thus, Brown did not belong to either the control or treatment group.
To Which Group Did Stanford Belong-Control or Treatment?
The F-value for the control group test was 19.16, significant at 0.00, and
the F-value for the treatment group test was 5.59, significant at 0.00.
This implied that Stanford did not belong to either the control or
treatment group.
To Which Group Did USC Belong-Control or Treatment?
We tested for which group USC belonged by excluding the SAT scores
variable (SCORESAT) from the model since the data were not available for
2003-2004. The F-value for the control group test was 23.23, significant
at 0.00, and the F-value for the treatment group test was 12.54,
significant at 0.00. This implied that USC did not belong to either the
control or treatment group.
Based on the above analysis, we determined that the best data for our
analysis was sample 2, and we excluded all five schools that only attended
the 568 Group meetings but did not implement the consensus approach.
3The test was not performed for Case Western Reserve and Dartmouth because
they are in samples 4 and 3, respectively. Samples 3 and 4 cannot be used
because there are no data for 1995-1996 and 2003-2004, respectively.
Appendix IV: Comments from 568 Presidents' Group
Appendix V: Consultants and Peer Reviewers
Hashem Dezhbakhsh, Ph.D. Professor of Economics
Emory University
Dennis Epple, Ph.D.
Thomas Lord Professor of Economics
Graduate School of Industrial Administration
Carnegie Mellon University
Janet Netz, Ph.D.
Founding Partner
ApplEcon LLC
Richard Romano, Ph.D.
Gerald L. Gunter Professor of Economics
Department of Economics
Warrington College of Business
University of Florida
Lawrence White, Ph.D.
Arthur E. Imperatore Professor of Economics
Department of Economics
Leonard N. Stern School of Business New York University
Gordon C. Winston, Ph.D.
Orrin Sage Professor of Political Economy, Emeritus
Director of the Williams Project on the Economics of Higher Education
Department of Economics
Williams College
Appendix VI: GAO Contact and Staff Acknowledgments
GAO Contact
Cornelia M. Ashby, Director, (202) 512-7215
Staff Acknowledgments
The following individuals made important contributions to the report:
Sherri Doughty, Assistant Director; Andrea Sykes; John A. Karikari; Angela
Miles; Daniele Schiffman; John Mingus; Dayna Shah; Richard Burkard; Susan
Bernstein; Rachel Valliere; Robert Alarapon; Thomas Weko; and L. Jerome
Gallagher.
Bibliography
Avery, C. and C. Hoxby."Do and Should Financial Aid Packages Affect
Students' College Choices?" National Bureau of Economic Research Working
Paper, No. 9482. 2003.
Bamberger, G., and D. Carlton."Antitrust and Higher Education: MIT
Financial Aid (1993)." Case 11. The Antitrust Revolution (Third Edition:
1993).
Carlton, D., G. Bamberger, and R. Epstein. "Antitrust and Higher
Education: Was There A Conspiracy to Restrict Financial Aid?" RAND Journal
of Economics, vol. 26, no. 1 (Spring 1995): 131-147.
Epple, D., R. Romano, S. Sarpca, and H. Sieg. "Profiling in Bargaining
Over College Tuitions," unpublished paper. January 21, 2005.
Hill, C., and G. Winston. "Access: Net prices, Affordability, and Equity
At a Highly Selective College." unpublished paper. December 2001.
Hill, C., G. Winston, and S. Boyd. "Affordability: Family Incomes and Net
Prices at Highly Selective Private Colleges and Universities." The Journal
of Human Resurces, vol. XL, no. 4 (2005): 769-790.
Hoxby, C. "Benevolent Colluders? The Effects of Antitrust Action on
College Financial Aid and Tuition." National Bureau of Economic Research
Working Paper, No. 7754. June 2000.
Kim, M. "Early Decision and Financial Aid Competition Among Need-Blind
Colleges and Universities." unpublished paper. May 1, 2005.
Morrison, R. "Price Fixing Among Elite Colleges and Universities," The
University of Chicago Law Review, vol. 59 (1992): 807-835.
Netz, J. "Non-Profits and Price-Fixing: The Case of the Ivy League."
unpublished paper. November 1999.
Netz, J. "The End of Collusion?: Competition After Justice and the Ivy
League Settle." unpublished paper. Fall 2000.
Salop, S., and L. White. "Antitrust Goes to College," Journal of Economic
Perspectives, vol. 5, no. 3 (Summer 1991): 193-2002.
Shepherd, G. "Overlap and Antitrust: Fixing prices in a Smoke-Filled
Classroom," The Antitrust Bulletin. Winter (1995): 859-884.
Winston, G., and C. Hill. "Access to the Most Selective Private Colleges
by High-Ability, Low-Income Students: Are They Out There?" unpublished
paper. October 2005.
(130463)
GAO's Mission
The Government Accountability Office, the audit, evaluation and
investigative arm of Congress, exists to support Congress in meeting its
constitutional responsibilities and to help improve the performance and
accountability of the federal government for the American people. GAO
examines the use of public funds; evaluates federal programs and policies;
and provides analyses, recommendations, and other assistance to help
Congress make informed oversight, policy, and funding decisions. GAO's
commitment to good government is reflected in its core values of
accountability, integrity, and reliability.
Obtaining Copies of GAO Reports and Testimony
The fastest and easiest way to obtain copies of GAO documents at no cost
is through GAO's Web site ( www.gao.gov ). Each weekday, GAO posts newly
released reports, testimony, and correspondence on its Web site. To have
GAO e-mail you a list of newly posted products every afternoon, go to
www.gao.gov and select "Subscribe to Updates."
Order by Mail or Phone
The first copy of each printed report is free. Additional copies are $2
each. A check or money order should be made out to the Superintendent of
Documents. GAO also accepts VISA and Mastercard. Orders for 100 or more
copies mailed to a single address are discounted 25 percent. Orders should
be sent to:
U.S. Government Accountability Office 441 G Street NW, Room LM Washington,
D.C. 20548
To order by Phone: Voice: (202) 512-6000 TDD: (202) 512-2537 Fax: (202)
512-6061
To Report Fraud, Waste, and Abuse in Federal Programs
Contact:
Web site: www.gao.gov/fraudnet/fraudnet.htm E-mail: [email protected]
Automated answering system: (800) 424-5454 or (202) 512-7470
Congressional Relations
Gloria Jarmon, Managing Director, [email protected] (202) 512-4400 U.S.
Government Accountability Office, 441 G Street NW, Room 7125 Washington,
D.C. 20548
Public Affairs
Paul Anderson, Managing Director, [email protected] (202) 512-4800 U.S.
Government Accountability Office, 441 G Street NW, Room 7149 Washington,
D.C. 20548
www.gao.gov/cgi-bin/getrpt? GAO-06-963 .
To view the full product, including the scope
and methodology, click on the link above.
For more information, contact Cornelia Ashby at (202) 512-7215 or
[email protected].
Highlights of GAO-06-963 , a report to congressional committees
September 2006
HIGHER EDUCATION
Schools' Use of the Antitrust Exemption Has Not Significantly Affected
College Affordability or Likelihood of Student Enrollment to Date
In 1991 the U.S. Department of Justice sued nine colleges and
universities, alleging that they had restrained competition by making
collective financial aid determinations for students accepted to more than
one of these schools. Against the backdrop of this litigation, Congress
enacted a temporary exemption from antitrust laws for higher education
institutions in 1992. The exemption allows limited collaboration regarding
financial aid practices with the goal of promoting equal access to
education. The exemption applies only to institutional financial aid and
can only be used by schools that admit students without regard to ability
to pay.
In passing an extension to the exemption in 2001, Congress directed GAO to
study the effects of the exemption. GAO examined (1) how many schools used
the exemption and what joint practices they implemented, (2) trends in
costs and institutional grant aid at schools using the exemption, (3) how
expected family contributions at schools using the exemption compare to
those at similar schools not using the exemption, and (4) the effects of
the exemption on affordability and enrollment. GAO surveyed schools,
analyzed school and student-level data, and developed econometric models.
GAO used extensive peer review to obtain comments from outside experts and
made changes as appropriate.
Twenty-eight schools-all highly selective, private 4-year
institutions-formed a group to use the antitrust exemption and developed a
common methodology for assessing financial need, which the group called
the consensus approach. The methodology used elements already a part of
another need analysis methodology; schools modified this methodology and
reached agreement on how to define those elements. By the 2004-2005 school
year, 25 of 28 schools in the group were using the consensus approach.
Schools' implementation of the approach varied, however, with officials
from 12 of the 25 schools reporting that they partially implemented it, in
part because they believed it would be costly to do so.
Over the last 5 years, tuition, room, and board costs among schools using
the antitrust exemption increased by 13 percent compared to 7 percent at
all other private 4-year schools not using the exemption. While the amount
of institutional aid at schools using the exemption also increased-it did
so at a slower rate. The average institutional grant aid award per student
increased by 7 percent from $18,675 in 2000-2001 to $19,901 in 2005-2006.
There was virtually no difference in the amount students and their
families were expected to pay between schools using the exemption and
similar schools not using the exemption. While officials from schools
using the exemption expected that students accepted to several of their
schools would experience less variation in the amount they were expected
to pay, GAO found that students accepted to schools using the exemption
and comparable schools not using the exemption experienced similar
variation in the amount they were expected to pay. Not all schools using
the consensus approach chose to adopt all the elements of the methodology,
a factor that may account for the lack of consistency in expected family
contributions among schools using the exemption.
Based on GAO's analysis, schools' use of the consensus approach did not
have a significant impact on affordability-the amount students and
families paid for college-or affect the likelihood of enrollment at those
schools to date. While GAO found that the use of the consensus approach
resulted in higher amounts of need-based grant aid awarded to some student
groups compared to their counterparts at schools not using the consensus
approach, the total amount of grant aid awarded was not significantly
affected. It was likely that grant aid awards shifted from non-need-based
aid, such as academic and athletic scholarships, to aid based on a
student's financial need. Finally, implementing the consensus approach did
not increase the likelihood of low-income or minority students enrolling
at schools using the consensus approach compared to schools that did not.
The group of schools using the exemption reviewed this report and stated
it was a careful and objective report. However, they had concerns about
the data used in GAO's econometric analysis, which GAO believes were
reliable.
*** End of document. ***