PRELIMINARY, PLEASE DO NOT QUOTE OR CITE. Prices and Market Segments of Four Year Colleges

Transcription

1 PRELIMINARY, PLEASE DO NOT QUOTE OR CITE Prices and Market Segments of Four Year Colleges Amy Ellen Schwartz Wagner School of Public Service New York University Benjamin P. Scafidi Andrew Young School of Policy Studies Georgia State University This work is partially funded under a contract with the Bureau of Labor Statistics (BLS). Any findings or views expressed in this paper are the authors' and do not necessarily represent the views of BLS.

2 I. Introduction Since the earlier 1980 s the sticker price or list price of a college education in the United States has, according to estimates from the Consumer Price Index, risen significantly faster than the overall rate of inflation. This has raised considerable concern among policymakers, parents and students that college attendance was becoming less and less affordable even as it was becoming more and more important for economic success in the job market. Interestingly, for the CPI, the government collects data on the sticker price of college (tuition and fees) without adjusting for scholarships given or other discounts. No adjustments are made for changes in the quality or characteristics of the services provided, such as attributes of the faculty, the course offerings, or the facilities. Thus, the estimated price indices reflect changes in quality and characteristics of college as well as changes in prices. Furthermore, all four-year colleges are viewed as the same good. Clotfelter (1996) and others suggest that elite colleges are in different markets than other institutions. Among other colleges, it is possible that there are additional, but related, market segments as well. For example, are highly selective colleges, other four-year institutions, and junior colleges the same good or related goods with prices determined in separate, but related, markets? Consumers preferences for attributes of college and/or price increases may not be uniform across market segments. Triplett (1971) describes a method to determine whether two goods should be treated as separate products: Suppose we have two sets of observations, Y 1 and Y 2, each set consisting of a number of varieties of a product. (footnote: For example, Y 1 might be a sample of station wagons, Y 2 of pickup trucks.) We wish to know if both sets belong to the same product classification, or if they should instead be treated as separate products. Suppose further that we have found a set of quality characteristics X 1, X k, for set Y 1. Then we propose the following rule: Y 1 is the same product as Y 2 if the set of characteristics X 1, X k, for Y 1 are also characteristics for Y 2, and if the implicit prices (the coefficients estimated from the hedonic quality function) for the set of characteristics are the same. Empirically this amounts to a test of whether the same quality function can be fittted to both sets of data, and translated to a relatively standard statistical test. (pg. 35) 1

3 In this paper, we develop and explore the construction of quality-adjusted price indices for US colleges, based on the estimation of hedonic models of the consumer price of college hedonic models for all four-year colleges and for various (potential) segments of the four-year higher education market. While many studies have considered the underlying causes of the growth in tuition, these have typically focused on investigating changes in the costs of higher education, the sources of revenues (especially government support and private contributions), or more generally on estimating aspects of the supply or demand for college education. There has been, to our knowledge, no previous work estimating quality-adjusted price indices and hedonic equations for (potentially) different segments of higher education. 1 To estimate the hedonic price models, we use data from the Annual Survey of Colleges (ASC) from the College Board for the years 1989 to These include data on tuition and fees, financial aid and scholarships, characteristics of the student body (including test scores, demographics, enrollment), and attributes of the services provided by the college (including faculty characteristics, computer information, course offerings, etc.) The purpose of this study is to identify quantifiable measures of quality of higher education services; to develop a methodology for quality-adjusting prices for changes in the attributes of college; to test for the presence of different market segments; and to provide an empirical demonstration of the proposed methods of quality adjustment and testing for the presence of market segments. 2 Many citizens, policymakers, and researchers are concerned about access to higher education for low income and historically disadvantaged individuals 1 Articles by Charles T. Clotfelter and Gordon C. Winston and others in a Symposium on the Economics of Higher Education in the Winter 1999 Journal of Economic Perspectives pp provide a nice overview and introduction to the field. 2 Although they are important and interesting, this paper does not attempt to answer the following questions: What is the value of college to society? What are the returns to college for a student? Or what is the full cost of providing a college education? This paper measures the net out-of-pocket costs of college to consumers and the shadow prices that consumers face for different attributes of college. 2

4 (Kane, 1999; Bowen and Bok, 1997). In designing policies to promote access for these individuals to different segments of the higher education market, it is important to know what is happening to the consumer price of college in the various market segments. The next section contains a simple economic model of college choice, and section III describes a hedonic model based on the economic model. Section IV contains a data description, and section V the estimates of the hedonic models. Conclusions and future research are in section VI. II. Economic Model This section describes a simple model of college choice. The choice of whether to go to college or whether to attend a two-year college is outside the model. A household sends one of its members to a four-year college to obtain certain college outputs. Similar to Ehrenberg (1998), we classify the outputs of college as: - Education - Food, accommodations, and amusements - Minor league professional athletics - Research - Investment management. Complementary to Eherenberg s framework, college admissions officers like to say that consumers choose which college to attend based on resources and reputation. College, therefore, is a discrete commodity that has many attributes that each consumer values. The prices of each individual attribute are not observed to the consumer. The price of the bundle of attributes, the out-of-pocket tuition cost to the consumer, is observed. Using these insights, we construct a model of a household s choice of a college, conditional on attendance at a four-year college. There is an issue whether the consumer is the 3

5 student or the student s parents. Abstracting away from this issue, we model the consumer as the household. 3 Assume that consumers value two attributes of college, X 1j and X 2j. X 1j and X 2j are observed to each consumer. 4 C ij is the consumption of all other goods if the student from household i attends college j, Y i is household income, T j is the sticker price of tuition at college j, and A ij is the aid offered to the student from household i at college j. 5 Let U(X 1j,X 2j,C ij ) be the utility that household i derives from sending a member of its household to college j. Conditional on college attendance, household i will choose to send a member of its household to college j if 0) U(X 1j,X 2j,C ij ) > U(X 1k,X 2k,C ik ), k j, where C ij = Y i (T j A ij ). Each consumer observes its net price of sending its student to college j, (T j A ij ), but the prices of individual attributes of college are unobserved. Thus, in this model, each consumer chooses 3 Whether the student or her parents is the primary decision maker will likely affect a household s willingness to pay for various attributes of college. Since this paper seeks to estimate the market value of those attributes and because the primary decision maker is unobserved, in the empirical work we assume that the average decision making power among students and their parents is unchanged over the sample period. 4 In the empirical specification of the hedonic model in section III, some of the X s are observed and some are unobserved to the analysts. Since we have panel data, we will be able to allow for the presence of unobserved time invariant university attributes in the hedonic model through a fixed-effects specification. We will employ a fixed effects specification in future work. 5 Rothschild and White (1996) suggest two reasons why colleges offer financial aid (A). 1) Given high fixed costs and low marginal costs, colleges can maximize net revenues through price discrimination. 2) Students can be viewed as inputs in the production of education. For colleges that offer aid to maximize net revenues, Rothschild and White believe that financial aid can be correctly viewed as a price discount. When financial aid is actually a payment to individual students for the inputs each provides, they believe that financial aid must be considered differently. In our model, financial aid that is a payment for inputs provided by individual students may also be viewed as a price discount from the consumer s point of view if student inputs have a market value of zero outside of college. From her own point of view, an individual student s input is a sunk cost. Inputs provided by students can be viewed as their presence at the college providing excellence and/or diversity, each of which may provide positive peer effects in the production of education. In deciding which college to attend, consumers thus care about the excellence and diversity of their peers (and all other attributes of college) and their own net price of college tuition minus aid. In our hedonic models described below, we include measures of excellence and diversity as attributes of college. 4

6 which college to attend based on the bundle of attributes (the X s) that each college provides and the net price that she faces (T-A). Suppose that for a particular attribute (X), du/dx > 0. Consumers would be willing to pay more for colleges that supply more of X, ceteris paribus. III. Hedonic Model Estimating a quality-adjusted price index for college tuition proceeds with a hedonic analysis of the price of college based, conceptually, on the work in Rosen (1974) and following, essentially, the methodology outlined in Triplett (1990), Berndt (1991) and others. Adopting a log-linear specification, the price of a year of school at the jth college at time t is written as a function of the characteristics of a year of college: (1) P jt = α + β Z Z j + β X X jt + β S S jt + ρ t I t + ε jt; j=1,.,j; t=1,,t. where Z j is a vector of time invariant characteristics of college j i.e., location, etc; X jt is a vector of time varying characteristics of college j at time t, i.e., size of the undergraduate student body, student/faculty ratio, computers per students, availability of dorms, course offerings, religious affiliation, quality of the faculty; S jt is a vector of time varying characteristics of the student body attending college j in time t, I t are a vector of year dummies variables that take a value of one in year t for t=1,, T; and P jt is the logarithm of the average price of a year at college j in time t. Since few students actually pay the list price for college because they receive at least some financial aid, we define P jt as the net or discounted price: 5

7 (2) P jt = T jt - A jt where T jt is the tuition (plus fees) price for one year for one undergraduate student (full-time) at college j in time t, (the sticker price or list price) and A jt is average financial aid for one year at college j in time t. Financial aid was restricted to grants only thus, student loans and workstudy income were not included. 6 As usual, equation (1) can be estimated to yield estimates of the marginal impacts of the characteristics of the colleges on their prices (βs). Each β is an estimate of the shadow price to consumers of a particular attribute of college. The coefficients ρ can be used to form a price index. Normalizing the level of the quality adjusted college price index to 100 in t, estimates of the price index for the following years can be created by exponentiating the ρ s. For example, the quality adjusted price index for t + 1 can be found as 100*exp(ρ t+1 ); the quality adjusted price index for t + 2 can be found as 100*exp(ρ t+2 ). Equation (1) forms the centerpiece of our empirical work. Note that although we estimate equation (1) essentially as specified, an interesting alternative would be to transform it using first differences or a similar transformation. The advantage is that this will reduce any bias due to the omission of unobserved time-invariant characteristics. The disadvantage is that all time-invariant characteristics will be eliminated from the analysis and the coefficients identified only by the variation across years. While the econometric strategy is relative straightforward, estimation presents a host of conceptual and practical questions. In particular, what are the relevant characteristics that need to 6 The College Board s ASC data include measures of average student loans and average income from work-study jobs. We did not include these variables in the financial aid measure. Ideally, we would include measures of the value of student loan terms and the value of work study jobs that are better than could be obtained in private capital 6

8 be included in the analysis? (How can/should those characteristics be measured?) Further, how should the sample of colleges be defined? (Only four year colleges? Two-year schools as well? Etc.) Are public and private colleges in the same market? Are colleges in different Carnegie classifications in the same market? To some extent, the answers depend upon the availability of data. IV. Data We use nine years of data from the College Board s Annual Survey of Colleges for the academic years to to estimate hedonic models following 1). The explanatory variables of a hedonic model of college should include only attributes of college that consumers value. This next subsection discusses the characteristics of college, and the subsection that follows describes our data. The Characteristics of College In principle, there is a long list of characteristics that would be necessary to fully describe the services (education; food, accommodations, and amusements; minor league professional athletics) provided by a college in exchange for a year s (discounted) tuition. These generally fall into four categories characteristics of the instructional program and student body (peers), physical characteristics of the school and other non-academic amenities, institutional/organizational characteristics of the school, and the value-added to each student s human capital. or labor markets. If one of these forms of financial aid has increased (decreased) over time in terms of its scope and/or generosity, then our methodology will overstate (understate) increases in the net price of college. 7

9 The Instructional Program. These include a broad range of characteristics and qualities of the instructional program - Average class size, number of seminars, course offerings (i.e., languages, labs, etc.), required (distribution) courses, pedagogical techniques, faculty characteristics (i.e., percent with Ph.D., percent with MA, number of Nobel prize winners?), percent of undergraduate courses taught by graduate students, publications of the faculty. The Student Body. The characteristics of the study body include a broad range of potential variables describing the students academic ability (i.e., average SAT scores, high school grades) and the students socioeconomic backgrounds (race/ethnicity, age, sex), as well as their interests and abilities in non-academic areas. School Characteristics Physical Characteristics. These include physical characteristics (such as dorms or gyms) and environmental characteristics (i.e., location, climate, proximity to major city, or important amenities like beaches or ski slopes). School Characteristics Institutional Attributes. These should also include institutional characteristics such as their religious affiliation, accreditation, the presence of other schools, participation in an athletic conference, whether they are public or private colleges, and if public whether they are state or local colleges. What about the value-added to each student s education? Surely, many of the attributes of college listed above contribute to the amount of education one receives from college. Verry and Davies (1976) described college as providing the following education outputs: (i) Instructional or teaching outputs (the transmission of knowledge). This involves the teaching of various kinds (general, vocational etc.), in different subjects and at different levels, all generally leading to certification of some description. (ii) Research outputs (the extension of knowledge). (iii)general Social Services. This is something of a catch-all category for the less tangible (and often most controversial activities of the university. It is intended to include the 8

10 general socialization function (the instillation of desirable work habits, co-operative behaviour, respect for laws and institutions, and, some would say, docility and obedience), and the related function, primarily benefiting employers, of sorting, selecting and screening individuals (page 10.) It is important to note that in paying for college, the consumer of undergraduate education buys only output (i). The college student (or her parents) is not buying research outputs (ii) or general social services (iii) directly with the price of tuition the price of tuition does not entitle you to (ii) or (iii), which may be available to some other community. Thus research and general social services will be important in the hedonic analysis only to they extent they enhance the undergraduate college experience. For example, research may enhance the reputation and/or learning experience at a particular college. Since colleges are multi-product firms, we cannot use the price indices estimated in this paper to deflate total university expenditures to get a measure of university output. For example, the benefits of research an important university output valued by non-students may be only partially capitalized in the net price of an undergraduate education. Obtaining measures of the value of all the outputs of a university and the full cost of providing those outputs is exceeding difficult (Winston 1999; Clotfelter, 1999) and is not the subject of this paper. More recently, some have argued that the output of a college is, at least to a significant degree, measured by the increase in the student s future earnings or their market value. Thus, some would include a measure of value added in the specification of the hedonic model. While calculations about the impact of a school on a child s future earnings is clearly an important consideration for students and parents choosing schools, it is unclear how those calculations are made given the available data. Put simply, we know of no broadly available data on the salaries 9

11 of recent (or other) graduates, or any other direct measure of the value-added to a student s education, for individual American colleges for each of the last ten years. 7 The Sample The College Board s ASC data includes information on: (1) institutional aspects of colleges, for example, source of control (i.e., private vs. public), religious affiliation or accreditation; (2) environment (i.e, urban vs. rural ) (3) facilities such as library holdings, availability of dorms (3) enrollment (part time, full time, etc.) (4) academic offerings and policies (5) fields of study (6) placement and credit policies (7) freshman admissions/profile (8) transfer study policies (9) Student life (sororities, intercollegiate sports, etc) (10) tuition and fees and (11) Financial Aid. These specific survey items vary somewhat over the sample, however, many elements remain the same over time. Table 1 provides descriptive statistics for the variables that we include in our model. Our analysis sample includes only 4-year colleges, that reported non-zero enrollment and expenditures, with at least 20% of the student body undergraduates (following Winston JEP). In addition, they must have a Carnegie classification and must be located in the 50 US states or Washington DC. There are missing values for particular data elements in some observations. Of the 16,374 cases over the 9 years, that were eligible to be in the sample, 12,579 had good data on tuition and fees, financial aid, Carnegie classification, and number of full and part-time students. 8,225 out 7 Literally, the relevant measures of value-added for a student considering to enroll in a particular college at time t would be the value-added to students who graduated at t-1. Presumably, prospective students would have better information about the value-added that individual schools provided to its recent graduates, than the value-added 4-5 years in the future. 10

12 of these 12,579 observations had good data on all variables included in the estimation. Computers per student and percent minority enrollment were the most frequently missing. Note that observations may be missing for a variety of reasons. Most straightforward is that a college may have failed to return the data in a given year. Colleges have an incentive to provide data, since the College Board provides that data to high school seniors shopping for schools, however, that incentive may be more important for some schools than others. Alternatively, a school may enter or exit because they fail to meet the criteria for inclusion in the sample schools offering only a two year degree in the early part of the study period may have offered four year degree in the end, thus entering the sample. Or, schools may have exited the 4 year market. This entry and exit is infrequent in terms of the number of students served by these schools. Given the missing data and entry and exit, we estimate the hedonic model with an unbalanced panel of colleges. V. Results We performed ordinary least squares (OLS) on various hedonic specifications of the price of college, each consistent with equation (1). Table 2 contains baseline hedonic regression results. As shown in specification (1) in table 2, a regression of the log of tuition and fees on the year dummies yields a monotonic yearly increase in the sticker price of college. Although a regression of the log of the net price of college [tuition plus fees minus aid specification (2)] on the year dummies shows a monotonic annual increases in the net price of college, the increase is lower over the time period. However, in the period , the net price of college was rising faster than the sticker price. From 1993 to 1997, sticker price increases were far greater than increases in net college prices resulting in the overall lower inflation in net prices relative to sticker prices by This lower inflation over the time period indicates 11

13 that while the sticker price of college was increasing rapidly between 1989 and 1997, rises in average financial aid per student which increased rapidly between 1993 and 1997 lead to lower inflation in the net price of college relative to the sticker price. The R**2 in each regression in table 4 is quite low and.031, respectively. Table 3 contains the results of an OLS regression of the log of net college price (tuition plus fees minus aid) on year dummies and several attributes of college. Estimates of the coefficient on each attribute are interpreted as their shadow market prices. In specification 3 in table 3, the coefficients on the year dummy variables increase monotonically. Most coefficients on the college attributes have the expected signs. Colleges with larger class sizes, more parttime students, locations outside the northeastern U.S., open admissions, greater percentages of minority students, and older entering freshman have lower net market prices, ceteris paribus. 8 Colleges that offer engineering majors, Ph.D. and masters degrees, higher SAT scores of entering freshman, and dormitories have higher market net prices. 9 Colleges that are older, private, members of the NCAA, have division 1 football, and located in larger cities have higher net prices, as well. Some variables did not have the expected sign, but these counterintuitive signs may be due to multicollinearity. For example, while the coefficient estimate on the percent of faculty 8 The model considers percent of students from racial and ethnic minorities, the average age of entering freshman, and standardized test scores of freshman as attributes of college that are valued by consumers of college. In the regressions reported in tables 3 and 4, these three attributes are lagged. For example, the log of the net price charged for the academic year is a function of the percent of minority students enrolled in the academic year. Contemporaneous attributes of the student body (peers) may be correlated with the error term, and by using the lagged measures of these attributes we escape this problem. If marginal consumers truly choose which college to attend based on net prices and resources and reputation, then the prior year measures are the measures of college quality on which they base their enrollment decisions. 9 Some colleges reported SAT scores only, some reported ACT scores only, and some reported neither. The coefficient on ACT scores is negative, which seems counterintuitive. However, including a dummy variable for whether the school reported only ACT scores led to a positive coefficient on ACT scores and a negative coefficient on the ACT dummy colleges that require that students report ACT scores only have lower market net prices. The ACT and SAT variables in the regressions in tables 5 and 6 measure both the effect of higher scores and which test is required. 12

14 who are full-time is negative, the coefficient on percent of faculty who have Ph.Ds is positive and slightly larger. The Carnegie classification variables have interesting magnitudes: Liberal Arts I and Liberal Arts II colleges have higher net prices, ceteris paribus, than Research I and Research II colleges, respectively. Doctorate granting institutions have higher net prices than Comprehensive colleges. Each of the aforementioned Carnegie classifications have higher net prices than the excluded category, other Carnegie classifications. These other Carnegie classifications include art schools, music schools, business schools, seminaries, etc. Estimates from each of these three regression specifications are used to construct price indexes in table 5. These price indexes are compared to the CPI-U and the CPI for college tuition and fees. The CPI-U rose 29.4% between 1989 and 1998, while the CPI for tuition and fees rose 81.7%. Using our sample of four-year colleges, we estimate that tuition and fees rose 76.5% over the sample period. Adjusting for financial aid, our sample suggests that the net price of college increased by only 65.1% over the sample period. Quality-adjusting the price indexes leads to further changes in the estimated price index over this time period. The estimates from the quality-adjusted hedonic model for all colleges suggests that the net price of tuition increased by 49% from 1989 to Public Versus Private Colleges Are public and private colleges the same good? That is, do consumers of public colleges value college attributes the same as consumers of private ones? This question is answered by running separated hedonic regressions for public and private colleges specifications (4) and (5) in table 3. The results suggest that public colleges have experienced double the increase in net 13

15 prices as private colleges between Differences in the coefficients on attributes are large in magnitude as well. 10 There is a much larger premium to smaller classes in private colleges, but the effect of having a Ph.D. program is smaller in private colleges. Having a division 1 football team and fraternities/sororities increases the net price of private colleges, while decreases net prices of public colleges. The Carnegie classification variables have larger positive coefficients in the private regression, especially the coefficients on the liberal arts classifications. The age of the college has a large positive coefficient in the public regression. The presence of dormitories has a big positive effect on the net price of public colleges, but almost no effect on private colleges. The percentage of minority students has a larger negative effect on the net price of private colleges. Interestingly, SAT scores have about the same effect on net prices in both the public and private regressions. Table 5 shows the differences in the estimated price indices for public and private colleges. While the net price of public colleges increased by 81.7% over the sample period, the net price of private colleges increased by only 36.5%. Carnegie classifications Table 4 displays separate hedonic regression results for four of the Carnegie classifications: Research I, Comprehensive college, Liberal Arts I, and Liberal Arts II. The estimated price indices are in the bottom panel of table 5. Research I and Liberal Arts I institutions had the lowest rates on inflation, while Comprehensive colleges and Liberal Arts II had much larger negative coefficients on the percent of minority students. The age of the school had a much bigger effect for Research I and Comprehensive colleges. Liberal Arts I colleges 10 We have not yet performed the appropriate statistical tests to measure significance level of differences in estimated coefficients between regressions. 14

16 had much lower net prices the larger the class sizes. As expected, consumers of Liberal Arts I institutions have stronger tastes for smaller classes than do consumers of other types of schools. The positive effect of percent of faculty with a Ph.D. was much larger for Comprehensive colleges and Liberal Arts II colleges. VI. Concluding Remarks and Future Work Estimates of a price index for higher education are sensitive to financial aid consideration and quality adjustment. Unadjusted price indices, such as the ones used to date by the BLS, reflect changes in the quality of college education as well as changes in the price. In particular, our preliminary results suggest that quality adjusted price indices indicate less inflation in the price of college than other indices suggest. Public and private colleges seem to have different rates of price increase as well as different shadow prices for various attributes of college; the same holds true for schools with different Carnegie classifications. While the conceptual framework for implementing a quality-adjusted price index for higher education is straightforward, practical implementation presents empirical challenges identifying data sources for characteristic variables with consistent definitions, and with consistently available data. In our current work, we use an unbalanced panel of colleges to estimate our hedonic models. This could bias the estimates of the coefficients on the year dummies if the mix of schools changes year-to-year in a non-random way. One alternative would be to construct a balanced panel of colleges. This alternative suggests a trade-off between sample size and the richness of the characteristics included in the regressions. Many data elements are missing for particular schools or particular observations which means that 15

17 constructing a balanced panel over the nine year period can only be accomplished at the cost of a significantly reduced sample. In the future, we will explore alternative solutions including the following. First, where data for an explanatory variable data is missing, we can substitute interpolated data from adjacent years, or, more generally, use data from non-missing years to create approximations in available years. Second, we can get additional data. We plan to supplement the College Board data with additional data obtained from the US Department of Education Integrated Post- Secondary Education Data System (IPEDS). Another approach would be to use a fixed effects specification to control for unobserved heterogeneity, which eliminates all of the time-invariant characteristics as well. The interpretation of a fixed effects specification is appealing the fixed effects capture what is essentially a brand effect for each college in the sample, that is similar to the brand effects that might be included in a hedonic analysis of automobile prices. Other future work includes: - Thinking about the advantages and disadvantages of various weighting schemes in the regressions. - Incorporating 2-year colleges into the analysis. - Looking at the effect of competition within and across college submarkets on net prices Identifying best buys and comparing these to Barron s and other rankings. - Analyzing what exactly drives the difference between the adjusted and unadjusted price indices and the differences between price indices in different submarkets. - Examining the regional and/or spatial elements of the price of college. 11 Hoxby (1997) looks into this issue. 16

18 References Berndt, Ernst R. (1991) The Measurement of Quality Change: Constructing an Hedonic Price Index for Computers Using Multiple Regression Methods Chapter 4 in The Practice of Econometrics Classic and Contemporary Addison-Wesley:New York pp Berndt, Ernst. R. and Griliches, Zvi (1993) Price Indexes for Microcomputers: An Exploratory Study in Price Measurements and Their Uses Foss, Manser and Young, ed. National Bureau of Economic Research, p Bowen, Howard R. (1980) The Costs of Higher Education Jossey-Bass Publishers, San Fransisco. Bowen, William G. and Bok, Derek (1998) The Shape of the River: Long Term Consequences of Considering Race in University Admissions Princeton University Press, Princeton, NJ. Clotfelter, Charles T. (1996) Buying the Best: Cost Escalation in Elite Higher Education NBER, Princeton University Press, Princeton, NJ Clotfelter, Charles T. (1999) The Familiar but Curious Economics of Higher Education Introduction to a Symposium Journal of Economic Perspectives Winter p Clotfelter, Charles T. and Michael Rothschild (1993) editors Studies of Supply and Demand in Higher Education National Bureau of Economic Research, University of Chicago Press, Chicago Griliches, Zvi (1971) editor Price Indexes and Quality Change: Studies in New Methods of Measurement, Harvard University Press, Cambridge, MA Griliches, Zvi, editor (1992) Output Measurement in the Services Sector National Bureau of Economic Research, University of Chicago Press, Chicago Hauptman, Arthur M. with Jamie Merisotis (1990) The College Tuition Spiral The American Council on Education and the College Board Hoxby, Caroline M. (1997) How the Changing Market Structure of U.S. Higher Education Explains College Tuition, NBER Working Paper 6323, December Kane, Thomas (1999) The Price of Admission: Rethinking How Americans Pay for College Brookings Institution Press, Washington, D.C. Liegey, Paul R. (1993) Adjusting Apparel Indexes in the Consumer Price Index for Quality Differences in Price Measurements and Their Uses Foss, Manser and Young, ed. National Bureau of Economic Research p

19 Rosen, Sherwin (1974) Hedonic Prices and Hedonic Markets: Product Differentiation in Pure Competition, Journal of Political Economy, April, pp Rothschild, Michael and Lawrence White (1996) The Analytics of the Pricing of Higher Education and Other services in Which Customers are Inputs, Journal of Political Economy, June, pp Triplett, Jack E. (1971) The Theory of Hedonic Quality Measurement and Its Use in Price Indexes BLS Staff Paper 6, United States Government Printing Office, , Washington, DC Triplett, Jack (1990) Hedonic Methods in Statistical Agency Environments in Berndt, E.R. and J.E. Triplett (eds.) Fifty Years of Economic Measurement: The Jubilee Conference on Research in Income and Wealth, NBER Studies in Income and Wealth, Chicago: University of Chicago Press. Verry, Donald. and Bleddyn Davies (1976) University Costs and Outputs Elsevier, New York Winston, Gordon C. (1999) Subsidies, Hierarchy and Peers: The Awkward Economics of Higher Education Journal of Economic Perspectives Winter p

20 Variable Definitions Variable Tuition+Fees-Aid per student Tuition + Fees Aid Per Student Definition Average tuition & fees minus average aid per student; average tuition is the weighted mean of in-state and out-of-state tuition--weighted by number of students paying each tuition level Average tuition & fees; as computed above Average cash grants to students; does not include work-study aid or student loans Instructional Program LN_Pupil-Teacher ratio Log (full-time +.5*part-time students divided by the number of full-time plus part-time faculty) % Faculty with PhD Percent of full-time plus part-time faculty with PhD % Full-time Faculty Number of full-time faculty divided by (number of full-time plus part-time faculty) Business major offered =1, if business major offered; 0 otherwise Engineer maj. Offered =1, if engineering major offered; 0 otherwise Computers per student # of personal computers available for student use divided by full-time +.5*part-time students Student Body LN Full-time Students Log of full-time students LN Part-time Students Log of part-time students ACT 75th percentile 75th percentile of ACT score of freshman who enrolled in previous year; 0 if not reported SAT 75th percentile 75th percentile of SAT score of freshman who enrolled in previous year; 0 if not reported SAT/ACT Missing, test req. SAT and ACT Missing; and a standardized test is required SAT/ACT Missing, test not req. SAT and ACT Missing, standardized test not required SAT/SAT Missing, test req? SAT and SAT Missing, whether standardized test is required is missing % Minority enrollment Percent of student body classified as a racial or ethnic minority in previous year Average age freshman Average age of freshman who enrolled in previous year Open Admissions =1, if open admissions policy; 0 otherwise application acceptance rate number of applications accepted for admission divided by the number of applications received 19

21 Variable Definitions, continued Variable Institutional Characteristics Research University I Research University II Doctorate Granting U Comprehensive College Liberal Arts I Liberal Arts II PhD granting institution Masters' granting inst. Age of School Single Sex Private Southern Baptist Catholic Methodist Presbyterian Evangelical Lutheran Definition =1, if Carnegie classification is Research University I =1, if Carnegie classification is Research University II =1, if Carnegie classification is Doctorate Granting University I or II =1, if Carnegie classification is Comprehensive College I or II =1, if Carnegie classification is Liberal Arts I =1, if Carnegie classification is Liberal Arts II =1, if institutions grants PhDs; 0 otherwise =1, if institution grants Masters' degrees; 0 otherwise =1989 minus year school founded =1, if institution is single sex; 0 otherwise =1, if institution is private; 0 otherwise =1, if institution is Southern Baptist =1, if institution is Catholic =1, if institution is Methodist =1, if institution is Presbyterian =1, if institution is Evangelical Lutheran Location Characteristics Mid Atlantic =1, if institution is located in the mid-atlantic U.S. Southeast =1, if institution is located in the southeast U.S. Midwest =1, if institution is located in the midwestern U.S. Southwest =1, if institution is located in the southwestern U.S. West =1, if institution is located in the western U.S. Over 500,000 in city =1, if institution is located in a city with over 500,000 residents ,999 in city =1, if institution is located in a city between 250,000 and 499,999 residents ,999 in city =1, if institution is located in a city between 50,000 and 249,999 residents Non-academic Amenities Frats and/or Sororities Dorms Available Div 1 Football Div 1 Men's Basketball NCAA =1, if institution has fraternity and/or sororities; 0 otherwise =1, if institution has dorms available; 0 otherwise =1, if institution has Division 1 football; 0 otherwise =1, if institution has Division 1 men's basketball; 0 otherwise =1, if institution is a member of the National Collegiate Athletic association (NCAA); 0 otherwise Source: College Board's Annual Survey of Colleges (ASC)

22 Table 1 Summary Statistics N Mean Std. Dev. Min Max Tuition + Fees - Aid per student , , Tuition + Fees , , Log(Tuition+Fees-Aid per student) Instructional Program LN_Pupil-Teacher ratio % Faculty with PhD % Full-time Faculty Business major offered Engineer maj. Offered Computers per student Student Body LN Full-time Students LN Part-time Students ACT 75th percentile** SAT 75th percentile** SAT/ACT Missing, test req SAT/ACT Missing, test not req SAT/SAT Missing, test req? % Minority enrollment Average age freshman Open Admissions Institutional Characteristics Research University I Research University II Doctorate Granting U Comprehensive College Liberal Arts I Liberal arts II PhD granting institution Masters' granting inst Age of School Single Sex Private Southern Baptist Catholic Methodist Presbyterian Evangelical Lutheran

23 Table 1, continued N Mean Std. Dev. Min Max Location Characteristics Mid Atlantic Southeast Midwest Southwest West Over 500,000 in city ,999 in city ,999 in city Non-academic Amenities Frats and/or Sororities Dorms Available NCAA ** Schools that do not report this variable are assigned a value of 0. 22

24 Table 2 Hedonic Regression Results No Quality Adjustments (1) LN(Tuition+Fees) (2) LN_Net Price* Estimate Std. Error Estimate Std. Error Constant p p p p p p p p N 12,597 12,597 R** * Net Price equals tuition + fees - aid per student 23

25 Table 3 Hedonic quality-adjusted Regression Results Full sample Public colleges Private colleges (3) (4) (5) Estimate Std. Error Estimate Std. Error Estimate Std. Error Constant p p p p p p p p LN_Pupil-Teacher ratio % Faculty with PhD % Full-time Faculty LN Full-time Students LN Part-time Students Business major offered Engineer maj. Offered PhD granting institution Masters' granting inst Div 1 Football Div 1 Men's Basketball Research University I Research University II Doctorate Granting U Comprehensive College Liberal Arts I Liberal Arts II Age of School ( Founding) Single Sex Private Mid Atlantic Southeast Midwest Southwest West Over 500,000 in city ,999 in city ,999 in city Frats and/or Sororities NCAA Dorms Available Average age freshman % Minority enrollment open admissions Computers per student Southern Baptist Catholic Methodist Presbyterian Evangelical Lutheran ACT 75th percentile SAT 75th percentile SAT/ACT Missing, test req SAT/ACT Missing, test not req SAT/SAT Missing, test req? N 8,225 2,440 5,785 R**

26 Table 4 Hedonic quality-adjusted Regression Results by Carnegie Classification Research I Comprehensive Coll.Liberal Arts I Liberal Arts II (6) (7) (8) (9) Estimate Std. Error Estimate Std. Error Estimate Std. Error Estimate Std. Error Constant p p p p p p p p LN_Pupil-Teacher ratio % Faculty with PhD % Full-time Faculty LN Full-time Students LN Part-time Students Business major offered Engineer maj. Offered PhD granting institution Masters' granting inst Div 1 Football Div 1 Men's Basketball Age of School (1989-Founding) Single Sex Private Mid Atlantic Southeast Midwest Southwest West Over 500,000 in city ,999 in city ,999 in city Frats and/or Sororities NCAA Dorms Available Average age freshman % Minority enrollment open admissions Computers per student Southern Baptist Catholic Methodist Presbyterian Evangelical Lutheran ACT 75th percentile SAT 75th percentile SAT/ACT Missing, test req SAT/ACT Missing, test not req SAT/SAT Missing, test req? N 329 3, ,190 R** * Net Price equals tuition + fees - aid per student 25