Financial resources, regulation, and enrollment in US public higher education

Economics of Education Review 21 (2002) 101–110 www.elsevier.com/locate/econedurev

Financial resources, regulation, and enrollment in US public higher education Mark C. Berger a

a,*

, Thomas Kostal

b

Department of Economics, Gatton College of Business and Economics, University of Kentucky, Lexington, KY 40506-0034, USA b Institute of Public Sector Economics, Vienna University of Economics and Business Administration, A-1090 Vienna, Austria Received 1 June 2000; accepted 4 October 2000

Abstract While total financial resources for higher education have been rising, there has been a significant shift in the share of resources coming from tuition and fees and a decline in the share coming from state appropriations. We seek to understand the enrollment consequences of this shift and to explore policy options using the results of a two-stage least-squares model of the demand for and supply of enrollment in public higher education. We estimate the model using 1990–95 data for the 48 continental US states. Tuition, average wage levels, and average education levels significantly affect enrollment demand, while state appropriations, other revenue, number of institutions, and the level of regulation significantly affect enrollment supply. Our simulations of policy options illustrate the difficulty of maintaining enrollment levels in the face of tuition increases. If tuition continues to rise, states are faced with reducing supply through lower state appropriations, or attempting to maintain current supply by increasing the amount of regulation in higher education.  2002 Elsevier Science Ltd. All rights reserved. JEL classification: I2 Keywords: Educational economics; Educational finance; Demand for higher education; Supply of higher education

1. Introduction

Changes in the financing of higher education in the 1990s have gradually shifted the burden of paying from the state to the individual. Rapidly increasing costs […] combined with continued intense competition for state resources […] and enrollment jumps […] threaten to exacerbate those changes (Breneman & Finney, 1997, p. 55). Although the total financial support for higher education in the United States is continuously increasing —

* Corresponding author. Tel.: +1-859-257-1282; fax: +1859-257-7671. E-mail address: [email protected] (M.C. Berger).

measured as a share of gross domestic product (GDP), the higher education sector reached an all-time high of nearly 3% in 1995 — structural changes in the financial pattern of this support are striking. Especially for public colleges and universities the changes since the early 1980s are dramatic. Whereas there have been increases in tuition and fees as well as the share of support from “private gifts, grants, and contracts”, the share of support from state appropriations has been steadily decreasing. Overall, this development can be characterized as a significant shift in the financial burden from the public sector — the state governments — to the private sector — the students and their families. Should these trends continue, considerable consequences concerning supply and demand for public higher education are possible in the long run. If that is the case, what effects on policy objectives concerning higher education are to be expected? One of

0272-7757/02/$ - see front matter  2002 Elsevier Science Ltd. All rights reserved. PII: S 0 2 7 2 - 7 7 5 7 ( 0 0 ) 0 0 0 6 5 - 0

102

M.C. Berger, T. Kostal / Economics of Education Review 21 (2002) 101–110

the major objectives, especially for public institutions, is to provide broad access to higher education. However, shifting the financial burden from state government to students and their families appears to work against the objective of maintaining broad access to higher education. Currently, the policy responses to this financial development seem to be just short-term oriented or ad hoc in nature. Long-term oriented state policies are essential, in particular, to maintain or improve access to higher education programs and to secure affordability for the future (Breneman & Finney, 1997, pp. 51–52). For a more complete understanding of the various policy options, empirical evidence on both supply-side and demand-side effects of the recent financial developments is necessary. While there are many studies dealing with the demand for higher education (e.g., Becker, 1990; Heller, 1999; Leslie & Brinkman, 1987; McPherson & Schapiro, 1991; Wetzel, O’Toole, & Peterson, 1998) and there are some supply-side studies (e.g., Ehrenberg, Rees, & Brewer, 1993; McPherson & Schapiro, 1993), there are very few studies that consider both demand and supply forces (e.g., Hoenack & Pierro, 1990; Quigley & Rubinfeld, 1993). For public colleges and universities it is sometimes argued that the supply side does not matter when estimating the parameters of demand — and vice versa — because public institutions are not free to retain revenues from increased tuition charges and, typically, they have open admittance policies (Becker, 1990, pp. 162–163). However, this may no longer be the case in the wake of the recent financial developments described above. To shed light on these issues, this study analyzes the determinants of enrollment in public higher education, simultaneously considering both demand and supply forces. We pay particular attention to the extent to which differing financial resources, most notably tuition and state and local appropriations, influence the enrollment rate in public higher education in the United States. We estimate our demand and supply model for higher education using 1990–95 data for the 48 continental US states. Our estimates provide evidence on the magnitudes of the effects of various supply and demand forces on public higher education enrollment, and allow us to simulate the effects of various options facing policymakers in states in the US. The paper is divided into six sections. Section 2 provides an overview of the data used in the analysis. Section 3 describes the econometric model, particularly the demand and supply equations. In Section 4, the empirical results are presented and analyzed. In Section 5 we present the results of some simulations that show the likely enrollment changes caused by different policy options. A brief conclusion is found at the end of the paper (Section 6).

2. Data This study uses data for 48 continental states for the period from 1990 to 1995. Most of the data used in the analysis were obtained from the Digest of Education Statistics. Additionally, data from various surveys conducted by the US Bureau of the Census and from Volkwein and Malik (1997) are employed. All variables measured in money terms are inflation-adjusted by the 1995 consumer price index (CPI-U). The data show wide variation across states in the structure of higher education. For example, the public higher education enrollment rate varies from 26% (Massachusetts, 1990) to over 70% (Wyoming, 1990 and 1991). The dispersion in private enrollments is also considerable: under 1% in Nevada (1990–95) and over 40% in Massachusetts (1994 and 1995). States with low average tuition at public institutions are Texas, North Carolina, Wyoming, and Idaho, while comparatively high average tuition is found in Vermont, Pennsylvania, New Hampshire, and Massachusetts. State and local appropriations vary from a low of $750 (per 18- to 24-year-old population within the state) in New Hampshire (1990) to a high of over $3800 in Wyoming (1990). The descriptive statistics and a detailed description of the variables are provided in Table 1. To control for the size of the group most likely to attend institutions of higher education across states, the dependent variable enrollpub — as well as enrollpriv — is measured as the total enrollment in public (private for enrollpriv) fouryear and two-year universities and colleges relative to the 18- to 24-year-old population. To approximate the direct cost of higher education the statewide average on undergraduate tuition fees (“sticker prices”) at four-year public and private institutions, tuitionpub and tuitionpriv, are employed. For public colleges and universities this is the average for in-state students only. The variable wagediff is employed as a proxy for the wage differential between workers with and without a college or university degree. This variable is obtained from the Annual Survey of Manufactures and is the ratio of the wage of nonproduction workers to production workers in manufacturing. Because nonproduction and production workers do not perfectly match college and non-college graduates there is measurement error in this variable which, if classical, will bias the estimated effect on enrollment toward zero. Other possible sources of measurement error include different mixes of production and nonproduction workers across states and the fact that today’s wage differential may not reflect expected future returns. However, this measure does have the advantage of being available on a state-by-state basis over the sample period.1 Another possible source for this variable 1

One available validity check for the wagediff proxy is to compare it to national data on college⫺high school wage differ-

M.C. Berger, T. Kostal / Economics of Education Review 21 (2002) 101–110

103

Table 1 Descriptive statistics and variable descriptions Variable enrollpub tuitionpub

Standard deviation

Mean 44.13 2594

avgwage

25,391

tuitionpriv

10,073

income

33,045

wagediff unemp

1.68 5.92

educ

19.61

nonwhite urban

18.51 67.76

stateapprop

1885

otherrev

2027

facsalary enrollpriv instnumber admiflexFmed admiflexFhi acadflexFmed acadflexFhi

46,194 11.48 8.12 0.40 0.35 0.33 0.38

Sourcea

Description

9.45 Total enrollment in public institutions, % of population, ages 18 to 24 Average undergraduate in-state tuition of public four-year institution, 395 in real $ per year 3212 Average wages of production workers, in real $ per year Average undergraduate tuition of private four-year institution, in real $ 2927 per year 4934 Median household income, in real $ per year Ratio of wage of nonproduction workers to wage of production 0.18 workers 1.45 Total unemployment rate in % People with at least a bachelor’s degree, % of population, ages 25+ in 3.72 1990 11.84 Ratio of nonwhite to total population in % 14.53 Urban population in % in 1990 State and local appropriations, grants and contracts to public 513 institutions, in real $ per population, ages 18 to 24 “Other revenues” of public institutions in real $ per population, ages 818 18 to 24 Average salary of instructional faculty in public institutions, in real $ 5456 per year 8.07 Total enrollment in private institutions, in % population, ages 18 to 24 4.15 Number of public institutions per 100,000 persons, ages 18 to 24 – 1 if financial flexibility is medium – 1 if financial flexibility is high – 1 if academic flexibility is medium – 1 if academic flexibility is high

DES DES ASM DES CPS ASM SA SA PEP SA DES DES DES DES DES Volkwein Volkwein Volkwein Volkwein

a Sources: DES, Digest of Education Statistics, National Center for Education Statistics, US Department of Education (various years); ASM, Annual Survey of Manufactures, Bureau of the Census, US Department of Commerce; CPS, Current Population Survey, Bureau of the Census, US Department of Commerce; SA, Statistical Abstract of the United States, Bureau of the Census, US Department of Commerce (various years); PEP, Population Estimates Program, Bureau of the Census, US Department of Commerce; Volkwein, Volkwein and Malik (1997).

would be to calculate wage differences by state from the raw microdata from the Current Population Surveys. However, the CPS has small samples for many states which in some cases produces large swings in estimates from year to year in these states. Given these choices, we decided to use the Annual Survey of Manufactures data, especially since the Bureau of Census does not publish earnings by education level for all states given the lack of adequate sample sizes in some states. The variable otherrev includes all other revenues of public degree-granting institutions except tuition, state ences from the CPS. According to the 1997 US Statistical Abstract (Table 246), the mean earnings of workers with a bachelor’s degree relative to those with a high school diploma was 1.73 in 1995, which compares favorably with the mean of wagediff in our sample (1.68). While the variation in wagediff across the sample may contain measurement error, at least the mean is centered near the true mean using representative national data.

and local appropriations, grants, and contracts (i.e., particularly federal appropriations, grants, and contracts; private gifts, grants, and contracts; endowment income; and revenues of auxiliary enterprises and hospitals). To control for flexibility and state regulation of the public university sector, the dummy variables admiflexFmed, admiflexFhi, acadflexFmed, and acadflexFhi are employed. The data come from Volkwein and Malik (1997), who distinguish between two overall measures to analyze the flexibility and state regulation at public universities. The academic flexibility is a combination of six separate indicators (i.e., accountability requirements, degree requirements, departmental flexibility, program discontinuance flexibility, program flexibility, and academic standards flexibility). Administrative flexibility is a combination of five separate indicators (i.e., budget flexibility, budget detail, expenditure detail, revenue flexibility, and tuition and fee revenue flexibility).

104

M.C. Berger, T. Kostal / Economics of Education Review 21 (2002) 101–110

3. Model

approximation to the demand equation for the aggregate demand for public higher education is specified:

This study seeks to explain determinants of public higher education enrollment rates across states — particularly, the influence of state and local appropriations and tuition — at public colleges and universities under the changing financial framework in the 1990s. To this end an econometric demand and supply model is estimated simultaneously determining tuition fees and enrollments for public higher education.

enrollpubDit ⫽a1⫹a2tuitionpubit⫹a3avgwageit

3.1. The demand for higher education Usually demand studies attempt to estimate an income and a price effect — i.e., how changes in income and price affect college and university enrollment. However, such estimates can be influenced considerably by the choice and definition of the dependent and independent variables, the type of database, and the functional form of the underlying equation. In most studies, measures of tuition, tuition for substitute institutions (cross-price effects), labor market conditions, and financial aid and other socioeconomic indicators (for gender, income, ability, etc.) are employed as explanatory variables. Both grouped and individual data, either in a time-series or a cross-sectional analysis, are used (Becker, 1990). However, grouped cross-section time-series data are seldom used (Heller, 1999). In thinking about the nature of the commodity “higher education” it is useful to distinguish between a consumption and an investment good. Being a consumption good, the demand for higher education may vary with the own price, the prices of substitute commodities, and income. In theory, as income and the price of substitute education increase and as the own price falls, demand increases and vice versa. The own price consists of two components: the direct price (i.e., tuition and fees, etc.) and the indirect price, the opportunity cost (i.e., particularly, the loss of income while attending college or university). The investment motive for higher education is based on human capital theory which assumes that (higher) education enables students to become more productive workers with a higher earnings potential. Thus, (direct and indirect) costs of higher education (including current labor market conditions) and future earnings determine the demand for higher education. Lower current costs and a higher stream of future earnings would be associated with higher levels of enrollment. Most of the empirical studies combine these two motives. Therefore, the demand for higher education is a function of direct and indirect costs/prices (tuition and forgone earnings), prices of substitute education, income, and a proxy for the higher earnings potential from obtaining a college education. Using the 1990–95 data for the 48 continental states and variables described in Section 2, the following linear

⫹a4tuitionprivit⫹a5incomeit⫹a6wagediffit

(1)

⫹a7unempit⫹a8educit⫹a9nonwhiteit⫹a10urbanit ⫹eit, for i=Alabama, Arizona, …, Wyoming; and t=1990, 1991, …, 1995. Based on the usual assumptions underlying consumption theory, a2 should be negative. However, the sign of a3 is not clear a priori — a negative price effect may be offset by a positive income effect. In other words, avgwage may in part be standing in for income differences across states and over time. If private colleges and universities are substitutes, we would expect a4 to be positive. If income is reflecting income differences across households, we would expect a5 to be positive. Following the human capital hypothesis, and if wagediff reflects future earnings differences, then a6 should be positive. If unemp is a proxy for current labor market conditions, then higher unemployment rates should be associated with lower opportunity costs of enrolling in higher education, and according to human capital theory a7 should be positive. To control for state-to-state differences in “environmental” conditions for higher education, the variables educ, nonwhite, and urban are employed. We would expect a8 to be positive, with high values of educ reflecting either more taste for higher education or a higher return to higher education in a particular state in a given year. If there is a lower demand for higher education among nonwhites, in part perhaps due to discrimination, then a9 would be negative. On the other hand, if there is a greater demand for individuals with more education in cities, then a10 would be positive. 3.2. The supply for higher education While models of the supply of higher education are not as fully developed as models of demand, we can certainly specify a series of variables on which supply is likely to depend. For example, it is likely to be the case that the offering of spaces to students is some function of the available financial resources (i.e., in this study tuitionpub, stateapprop, and otherrev). However, this relation depends crucially on both the institution’s objectives and its autonomy. Perhaps one of the most important objectives of public colleges and universities is the improvement of the quality of education and, particularly, of the institution’s

M.C. Berger, T. Kostal / Economics of Education Review 21 (2002) 101–110

105

prestige and reputation (see e.g., McPherson & Schapiro, 1993). Available funds equal, it seems plausible that higher quality and/or prestige is correlated with lower enrollments. As a rough proxy for the quality and prestige of an institution, the average annual salary of the (instructional) faculty (facsalary) is employed. While the market for faculty is likely to be national, suggesting facsalary is a good proxy, salary levels in a given state may also depend in part on labor market conditions and amenity levels in that state. Public colleges and universities are subject to state and federal regulations. Volkwein and Malik (1997) distinguish between two combined indicators to measure the autonomy (flexibility) of the public university sector in a given state. These two qualitative measures are specified as dummy variables in this study (admiflexFmed, admiflexFhi and acadflexFmed, acadflexFhi) to control for state differences in the institution’s autonomy with respect to their academic and administrative decisions (see also the variable description in Section 2). The effect of increasing flexibility on enrollment is not clear a priori. On the one hand, it could be argued that an overall increase in flexibility leads to higher enrollment due to increased productivity at the institution level (i.e., colleges and universities are now better able to adjust to changing circumstances and needs). On the other hand, it could be argued that in the case of bureaucratic institutions like public colleges or universities, it is necessary to implement a tight regulation system to secure accountability and to avoid inefficiency. It is likely that an institution with comparatively loose regulations will devote more resources to activities such as nonfunded research. In this case, more autonomy would correspond to a lower enrollment rate. Either result is plausible and the true effects of administrative and academic flexibility must be determined empirically. Finally, to control for state-to-state differences in the scope of the higher education sector and the extent of private-sector alternatives, the variables instnumber and enrollpriv are employed. To model the number of places at public colleges and universities across states and over time, the following linear approximation to the aggregate supply equation is specified:

The usual supply-side arguments would suggest that the signs of the coefficients of the financial variables, b2, b3, and b4, should be positive. The effect of our proxy for the institution’s quality and prestige (b5) should be negative. As discussed above, the signs of b6 to b9 are theoretically uncertain and must be empirically determined. We would expect b10 to be positive (i.e., the higher the college and university density in a given state, the higher is the aggregate enrollment rate). Finally, strong private-sector alternatives in higher education should have a negative effect on public-sector enrollments; thus, the sign of b11 should be negative.

enrollpubitS⫽b1⫹b2tuitionpubit⫹b3stateappropit

In the second stage the endogenous variable tuitionpub in the demand and supply equations is replaced by its first-stage fitted values. The demand equation’s coefficients are identified by the following variables in the supply equation but omitted from the demand equation: stateapprop, otherrev, facsalary, admiflexFmed, admiflexFhi, acadflexFmed, acadflexFhi, instnumber, and enrollpriv. On the other hand, the supply equation’s coefficients are identified by variables in the demand equation but omitted from the supply equation: avgwage, tuitionpriv, income, wagediff, unemp, educ, nonwhite, and urban.

⫹b4otherrevit⫹b5facsalaryit⫹b6admiflexFmedit ⫹b7admiflexFhiit⫹b8acadflexFmedit ⫹b9acadflexFhiit⫹b10instnumberit⫹b11enrollprivit ⫹uit, for i=Alabama, Arizona, …, Wyoming; and t=1990, 1991, …, 1995.

(2)

4. Empirical results Eqs. (1) and (2) presented in the previous section are a part of a simultaneous-equation system with the following equilibrium condition: enrollpubSit⫽enrollpubDit ,

(3)

for i=Alabama, Arizona, …, Wyoming; and t=1990, 1991, …, 1995. Both equations of the structural model are overidentified, thus the two-stage least-squares method (2SLS) can be used to obtain unique estimates of the parameters of Eqs. (1) and (2) (Pindyck & Rubinfeld, 1998). In the first stage, the reduced-form equation explaining the endogenous variable tuitionpub is estimated using ordinary least squares (OLS): tuitionpubit⫽p1⫹p2avgwageit⫹p3tuitionprivit ⫹p4incomeit⫹p5wagediffit⫹p6unempit⫹p7educit ⫹p8nonwhiteit⫹p9urbanit⫹p10stateappropit ⫹p11otherrevit⫹p12facsalaryit⫹p13admiflexFmedit

(4)

⫹p14admiflexFhiit⫹p15acadflexFmedit ⫹p16acadflexFhiit⫹p17instnumberit⫹p18enrollprivit ⫹vit.

106

M.C. Berger, T. Kostal / Economics of Education Review 21 (2002) 101–110

To test for the strength of these instruments, “at a minimum, first stage F statistics should be reported” (Staiger & Stock, 1997). First, we test the joint hypothesis that all of the coefficients on the demand identifying instruments are simultaneously zero.

Table 2 2SLS estimates of the demand for higher education in the US a enrollpubD

Coefficient

Standard error

z

HD0: p10⫽p11⫽p12⫽p13⫽p14⫽p15⫽p16⫽p17⫽p18

tuitionpub avgwage tuitionpriv income wagediff unemp educ nonwhite urban constant

⫺0.00631** 0.00058** ⫺0.00030 ⫺0.00012 3.03250 ⫺0.06434 0.91969** ⫺0.11568 0.07642 26.52378**

0.00109 0.00020 0.00022 0.00010 2.92000 0.20138 0.19158 0.06202 0.04740 7.46884

⫺5.782 2.931 ⫺1.368 ⫺1.130 1.039 ⫺0.319 4.800 ⫺1.865 1.612 3.551

(5)

⫽0. The F-statistic is: F(9, 270)=34.9. Second, we test the joint hypothesis that all of the coefficients on the supply identifying instruments are simultaneously zero. H0S: p2⫽p3⫽p4⫽p5⫽p6⫽p7⫽p8⫽p9⫽0.

(6)

The F-statistic is: F(8, 270)=18.6. Both null hypotheses are rejected at the 1% level. Given the magnitude of these F-statistics, we can conclude that we do not have a problem with weak instruments and can proceed to the second stage of the 2SLS estimation. To estimate the second stage (i.e., to estimate the structural equations) the use of simple OLS is problematic because of the time series of cross-sections data sets. The typical problem with data sets of this type is that the behavior of the disturbances over the cross-section at a given point in time is likely to be different from the behavior of the disturbances of a given cross-sectional unit over time. Heteroskedasticity and cross-sectional correlation may be problems across states, while over time autocorrelation must be taken into account. For this study, an analysis of the residuals shows significant autocorrelation and a heteroskedastic error structure in the demand equation, and autocorrelation but no indication for heteroskedasticity in the supply equation. The “Stata Statistical Software” provides the appropriate options within the “xtgls” command. To control for autocorrelation an “AR(1)” process with a common correlation parameter for all panels is employed. Additionally, the assumption of heteroskedasticity underlies the demand equation (Stata statistical software, 1999). Further, since the number of time periods is small relative to the number of cross-sectional observations in our data, following Beck and Katz (1995) we report “panel-corrected standard errors” (PCSEs) in the second-stage estimation of the structural model. Table 2 presents the simultaneously determined estimates, standard errors, z-statistics, and statistical significance at the 1% or 5% level for the demand equation (1). The coefficient of the direct-cost variable tuitionpub is highly significant, has the expected sign, and is of a reasonable magnitude. The model predicts that a $100 increase in tuition at public colleges and universities leads to a decrease in the average enrollment rate of 0.63 percentage points (a survey on the dimensions of demand responses is provided by Becker, 1990; Leslie & Brinkman, 1987; McPherson & Schapiro, 1991). The avgwage variable has a positive, highly significant

a **, Significant at 1% level; *, significant at 5% level; 2SLS instrumented variable: tuitionpub; second stage: OLS with PCSEs; panels: heteroskedastic; time-series: AR(1) process with common coefficient=0.7837; number of observations=288 (48 states and 6 time periods); Wald chi2 (9)=97.23.

impact on the enrollment rate. A $1000 increase of production workers’ wages leads to a 0.58 percentage points higher enrollment rate. It appears that for public colleges and universities any negative time cost effect is offset by a higher positive income effect due to increased average earnings. In other words, avgwage is primarily capturing income differences across states and over time rather than differences in the time costs of attending college. It seems that private institutions are not a direct substitute for public institutions of higher education. The cross-price elasticity is not significantly different from zero. The result is similar for the income variable. Apparently income differences across states and over time not captured by avgwage, are not important for explaining enrollment differences. It could be that variation in average nonlabor income across states is not enough to identify any resulting enrollment differences. The estimated effect of the returns on investment in college, the wage differential between nonproduction and production workers (wagediff), has the expected sign but the coefficient is insignificant. Perhaps the time period is not long enough (1990–95) to capture the longterm increases in the returns to higher education that have been pointed out in the literature. For example, Katz and Autor (1999) report that while the earnings of college graduates relative to high school graduates in the US increased by approximately 17% from 1979 to 1995, the increase from 1990 to 1995 was less than 3%. This suggests that we are attempting to identify the effect of wagediff primarily from cross-sectional variation, which is a more difficult task. In addition, there may be measurement error in our proxy which, as discussed above, may bias the estimated coefficient toward zero. Thus, we would not want to conclude on the basis of

M.C. Berger, T. Kostal / Economics of Education Review 21 (2002) 101–110

our estimates that increases in the returns to higher education have no effect on the demand for higher education. Instead, given the plethora of previous studies on the subject and the potential measurement error in our proxy, we would tend to believe the positive effect reported in Table 2, even though it is statistically insignificant.2 The average labor market conditions in a given state, measured by the average unemployment rate (unemp), do not have a significant impact on the enrollment rate. Perhaps this is a result of the use of an overall (i.e., nonspecific) unemployment rate that does not adequately capture the potential labor market conditions for college students. Merz and Schimmelpfennig (1999) suggest a “degree-specific unemployment rate”. Unfortunately, such disaggregated unemployment rates are unavailable. All coefficients of the variables to control for state differences in “environmental” conditions are of the expected sign; however, only the coefficient on educ is statistically significant. An increase of one point in the percentage of the population with a bachelor’s degree leads to almost a one percentage point increase in the enrollment rate. The minority rate has only a weak impact at a lower significance level (10%). Turning to the supply equation (2), Table 3 shows its simultaneously estimated coefficients. Most of the coefTable 3 2SLS estimates of the supply of higher education in the US a enrollpubS

Coefficient

Standard error

z

tuitionpub stateapprop otherrev facsalary admiflexFmed admiflexFhi acadflexFmed acadflexFhi instnumber enrollpriv constant

⫺0.00183 0.00513** 0.00411** 0.00019 ⫺4.14813** ⫺4.81235** ⫺4.67350** ⫺2.72733* 0.48722** ⫺0.11858 25.01938**

0.00105 0.00136 0.00068 0.00010 1.24052 1.29477 1.15771 1.26057 0.12627 0.07473 4.383176

⫺1.742 3.776 6.036 1.875 ⫺3.344 ⫺3.717 ⫺4.037 ⫺2.164 3.859 ⫺1.587 5.708

a **, Significant at 1% level; *, significant at 5% level; 2SLS instrumented variable: tuitionpub; second stage: OLS with PCSEs; panels: homoskedastic; time-series: AR(1) process with common coefficient=0.7917; number of observations=288 (48 states and 6 time periods); Wald chi2 (10)=307.50.

2 Freeman (1986) provides a review of earlier studies. Berger (1988), Willis and Rosen (1979), and Zarkin (1985) all find that the decision to invest in a college education or in a particular field is significantly related to the earnings opportunities after making the investment relative to other opportunities.

107

ficients are highly significant and are of the expected sign. Tuition is not significantly related to the supply of places at public colleges and universities. While the estimated coefficient is negative, it is very small, and a simple OLS estimation of the supply equation (2) only shows a weak positive but insignificant impact. Perhaps it is not unreasonable to believe that supply is not significantly related to variation in average tuition. At many public universities, tuition makes up a small portion of overall funding, and in some cases (e.g., in the Commonwealth of Kentucky), public universities may not see tuition revenue directly but only through some complex formula. Both of the other financial variables, stateapprop and otherrev, have substantial, highly significant, positive impacts on enrollments. Both coefficients are of similar magnitudes. A $100 increase in state and local appropriations and other revenues (per capita for the population aged 18 to 24 years) leads to an increase in the enrollment rate of 0.51 and 0.41 percentage points, respectively. The impact of facsalary on enrollments is negative but not statistically significant at the 5% level. In addition, the magnitude of the estimated effect is quite small. A $1000 faculty salary increase per year corresponds to just a 0.2 percentage points increase in the enrollment rate. All coefficients of the dummy variables to control for institutional flexibility are statistically significant and negative. Higher administrative and academic flexibility are associated with lower public enrollment rates. For example, states with high administrative autonomy on average have four percentage points lower enrollment rates than states with a low administrative autonomy. These estimates seem to bear out the hypothesis that a college or university with relatively loose regulations can devote more resources to activities for which there is weak demand but which, nonetheless, are important to its staff (e.g., nonfunded research or providing spaces for relatively high cost graduate and professional students) (Hoenack & Pierro, 1990). The average college and university density in a given state has the expected positive impact on supply. One more public college or university per 100,000 persons aged 18 to 24 leads to almost half a percentage point increase in the enrollment rate. Although the coefficient of enrollpriv is statistically insignificant, it has the expected negative sign. A one percentage point higher total enrollment in private colleges and universities is correlated with a 0.12 percentage point lower enrollment rate in the public higher education sector.

5. Selected policy simulations Based on the empirical results we construct simulations to show the likely effects of selected policy changes

108

M.C. Berger, T. Kostal / Economics of Education Review 21 (2002) 101–110

on demand and supply in the higher education sector. The main purpose of these simulations is to improve the understanding of the relationship between policy changes and enrollments on the one hand, and the relative magnitudes of the effects of policy variables and other factors that determine enrollments on the other. Based on the estimated equations (Tables 2 and 3) we start our simulations at the mean values of all the variables across the data set (Table 1). We define the variables tuitionpub and stateapprop, and the dummy variables admiflexFmed, admiflexFhi, acadflexFmed, and acadflexFhi, as our policy variables. We assume that state governments can control these variables directly and consider the effects of changes in these and other variables on enrollment rates in public higher education. 5.1. Scenario 1: Higher tuition and constant state appropriations In the first scenario, we extrapolate the past trends in two of the policy variables. From 1990 to 1995 average tuition at public colleges and universities increased by 30% and state and local appropriations stayed almost constant, both measured in real terms. Thus, we assume an additional increase of tuition fees by 30% and constant state and local appropriations. This policy change leads to a demand-side reduction in the enrollment rate by five percentage points (from 44 to 39%). For enrollment to remain stable other factors must alter. For example, average wages of production workers, all other things equal, must increase by 34% (from $25,400 to $34,000), the average educational level must increase by 28% (from 19.6 to 25%), or the wage differential between skilled and unskilled workers must almost double (from 1.7 to 3.3). Alternatively, there would have to be some combination of changes in these variables such that enrollment stayed at a 44% level. This scenario illustrates the magnitude of the own price effect on enrollment decisions. In order to counteract this own price effect and maintain enrollment levels, the increases in average wage levels (income effect), education levels, or the returns to higher education that would be needed are staggering. This would even be the case if the effect of increases in the returns to higher education were understated by a factor of two due to measurement error. Thus, in this scenario, we have considered increases in tuition balanced by changes in other variables that allow enrollment levels to be maintained at current levels. If other variables such as income levels, education levels, or the returns to education do not increase enough, another way to maintain equilibrium in the face of the tuition increase and the resulting decline in demand, would be to decrease supply by a corresponding amount. Thus, in the next scenario, we consider what happens when supply decreases through a decline in state appropriations.

5.2. Scenario 2: Higher tuition and lower state appropriations In the second scenario we propose a simulation that combines the same 30% increase in public tuition as in scenario 1 with a large enough real reduction in state and local appropriations to balance the decrease in enrollment caused by the price increase. This occurs when the 30% increase in tuition is combined with a 30% decrease in state appropriations per individual aged 18 to 24 (i.e., from $1885 to $1319). The decline in state appropriations would lead to a decrease in supply from 44 to 41%. The remaining decrease on the supply side (41 to 39%) would come from the increase in tuition of 30% due to the negative estimated coefficient on tuitionpub in Eq. (2). In other words, not only is the price to students increasing in this scenario, but supply decreases through reductions in state appropriations such that the equilibrium enrollment level falls from 44 to 39%. However, it should be noted that while this maintains the equilibrium of supply and demand, it is inconsistent with the overall objective of maintaining current enrollment levels in public higher education. One way to reach equilibrium under this scenario while maintaining current enrollment levels would be to combine demand changes with changes in supply-side variables. Thus, not only would there have to be the demand-side changes as discussed in scenario 1, but also at least one (or some combination) of the following changes in supply factors: an increase in other revenues by 28% (from $2027 to $2595) or a doubling of university density (from 8.1 to 17 institutions per 100,000 inhabitants aged 18 to 24). Such large changes show how unlikely it would be that enrollment levels could be maintained in the face of both tuition increases and state appropriation decreases. 5.3. Scenario 3: State appropriations versus state regulation With scenario 3 we want to simulate the trade-off between state appropriations and state regulation in higher education. In particular, we would like to show how the supply of higher education reacts to various assumptions regarding state regulation. Again, we start our simulation at the mean values of all the variables across the sample. We compare three cases: 1. High overall regulation — i.e., variables admiflexFmed, admiflexFhi, acadflexFmed, and acadflexFhi equal zero; 2. High administrative regulation — i.e., variables admiflexFmed and admiflexFhi equal zero but acadflexFmed and acadflexFhi stay constant; 3. High academic regulation — i.e., variables acad-

M.C. Berger, T. Kostal / Economics of Education Review 21 (2002) 101–110

flexFmed and acadflexFhi equal zero but admiflexFmed and admiflexFhi stay constant. The proposed policy change in case 1 leads to a sharp increase in supply of public higher education. The enrollment rate rises from 44 to 50%. Thus, as our simulation shows, a high overall regulation would allow states (and local governments) to reduce their appropriations to colleges and universities by 61% (from $1885 to $730) in order for enrollments to remain unchanged. This simulation suggests that increases in regulation can be compensated for by considerable decreases in state appropriations and still maintain overall enrollment levels. Comparing cases 2 and 3 we find that regulation of administrative spheres — i.e., in particular, budgetary matters — has a greater effect on supply than regulation of academic spheres. Whereas in case 2 the enrollment rate increases by three percentage points (from 44 to 47%), in case 3 it increases by two percentage points (from 44 to 46%). In case 2, this would allow state and local appropriations to fall by 36%, and in case 3 by 27%, and still maintain constant enrollment levels.

6. Conclusion In this study a simultaneous demand and supply model is estimated to explain enrollment in public higher education in the United States. Particular concern was paid to the impact of financial resources — tuition as well as state and local appropriations — on enrollments. Most of the empirical results bore out the underlying theoretical hypotheses. With regard to demand for public higher education, tuition proves to be the most significant variable: as tuition increases the enrollment rate decreases. However, the strongest impact on the level of the average enrollment rate comes from the already existing general educational level in a state. Foregone earnings, prices of private alternatives, income, and the unemployment rate have much smaller effects. Our proxy for the investment motive, the relative wage differential, has a sizable effect (but statistically insignificant) on enrollment. On the supply side, the enrollment rate is influenced positively by available funds (except tuition) and the relative size of the higher education sector within a certain state and negatively by the scope of autonomy of colleges and universities. Tuition, faculty salaries, and enrollments at private institutions have comparatively weak impacts on enrollments in the public higher education sector. The findings of this study have important policy consequences. Our policy simulations suggest — if one of the major goals for public colleges and universities is to maintain a high enrollment rate in higher education —

109

that it is important to ensure a reasonable level of tuition and a sufficient amount of state and local appropriations. If the latter is not possible — particularly due to tight state budgets — a more rigid regulation system may be an alternative policy option for the states. In states without such measures, absent continuing upward trends in the returns to college education, the enrollment rate in public higher education is likely to decline in the long run.

Acknowledgements The study was initiated during the research stay of Thomas Kostal at the University of Kentucky and was supported by a Fulbright Scholarship.

References Beck, N., & Katz, J. N. (1995). What to do (and not to do) with time-series cross-section data. American Political Science Review, 89 (3), 634–647. Becker, W. E. (1990). The demand for higher education. In S. A. Hoenack, & E. L. Collins, The economics of American universities (pp. 155–265). Albany, NY: State University of New York Press. Berger, M. C. (1988). Predicted future earnings and choice of college major. Industrial and Labor Relations Review, 41 (3), 418–429. Breneman, D. W., & Finney, J. E. (1997). The changing landscape. Higher education finance in the 1990s. In P. M. Callan, & J. E. Finney, Public and private financing of higher education. Shaping public policy for the future (pp. 30–59). Phoenix, AZ: Oryx Press. Ehrenberg, R. G., Rees, D. I., & Brewer, D. J. (1993). How would universities respond to increased federal support for graduate students? In C. T. Clotfelter, & M. Rothschild, Studies of supply and demand in higher education (pp. 183– 206). Chicago, IL/London. Freeman, R. B. (1986). Demand for education. In O. C. Ashenfelter, & R. Layard, Handbook of labor economics (Vol. 1). (pp. 357–386). Amsterdam/New York/Oxford/Tokyo: North-Holland. Heller, D. E. (1999). The effects of tuition and state financial aid on public college enrollment. The Review of Higher Education, 23 (1), 65–89. Hoenack, S. A., & Pierro, D. J. (1990). An econometric model of a public university’s income and enrollment. Journal of Economic Behavior and Organization, 14 (4), 403–423. Leslie, L. L., & Brinkman, P. T. (1987). Student price response in higher education. The student demand studies. Journal of Higher Education, 58 (2), 181–204. Katz, L. F., & Autor, D. H. (1999). Changes in the wage structure and earnings inequality. In (pp. 1463–1555). O. C. Ashenfelter, & D. Card, Handbook of labor economics, Vol. 3A. Amsterdam/Lausanne/New York/Oxford/Shannon/Singapore/ Tokyo: Elsevier. McPherson, M. S., & Schapiro, M. O. (1991). Does student

110

M.C. Berger, T. Kostal / Economics of Education Review 21 (2002) 101–110

aid affect college enrollment? New evidence on a persistent controversy. The American Economic Review, 81 (1), 309–318. McPherson, M. S., & Schapiro, M. O. (1993). The effect of government financing on the behavior of colleges and universities. In M. S. McPherson, M. O. Schapiro, & G. C. Winston, Paying the piper: Productivity, incentives, and financing in U.S. higher education (pp. 235–250). Ann Arbor, MI: University of Michigan Press. Merz, M., & Schimmelpfennig, A. (1999). The demand for higher education in Germany. Proceedings of the 1998 third international conference of the German socio-economic panel study users. Vierteljahreshefte zur Wirtschaftsforschung, 68(2), 204–208. Pindyck, R. S., & Rubinfeld, D. L. (1998). Econometric models and economic forecasts. (4th ed.). Boston, MA: McGraw Hill. Quigley, J. M., & Rubinfeld, D. L. (1993). Public choices in public higher education. In C. T. Clotfelter, & M. Rothsch-

ild, Studies of supply and demand in higher education (pp. 243–278). Chicago, IL/London. Staiger, D., & Stock, J. H. (1997). Instrumental variables regression with weak instruments. Econometrica, 65 (3), 557–586. Stata statistical software: Release 6.0. College Station, TX: Stata Corporation. Volkwein, J. F., & Malik, S. M. (1997). State regulation and administrative flexibility at public universities. Research in Higher Education, 38 (1), 17–42. Wetzel, J., O’Toole, D., & Peterson, S. (1998). An analysis of student enrollment demand. Economics of Education Review, 17 (1), 47–54. Willis, R. J., & Rosen, S. (1979). Education and self-selection. Journal of Political Economy, Part 2, 87 (5), S7–S36. Zarkin, G. A. (1985). Occupational choice: an application to the market for public school teachers. Quarterly Journal of Economics, 100 (2), 409–446.