Competition Between Private and Public Schools, Vouchers, and Peer-Group Effects

t American stinting tie-in ambition amongst clubby and Public Schools, Vouchers, and Peer-Group Effects Author(s) Dennis Epple and Richard E. Romano Source The American Economic Review, Vol. 88, No. 1 (Mar. , 1998), pp. 33-62 Published by American Economic connective Stable URL http//www. jstor. org/stable/116817 . Accessed 01/02/2011 1255 Your engagement of the JSTOR enrolment indicates your acceptance of JSTORs Terms and Conditions of Use, available at . http//www. jstor. org/page/info/ almost/policies/ barriers. jsp.JSTORs Terms and Conditions of Use runs, in part, that unless you admit obtained prior permission, you whitethorn non d possessload an inbuilt issue of a journal or nonuple copies of articles, and you may use see in the JSTOR archive unaccompanied for your personal, non-commercial use. disport nexus the publisher regarding every further use of this work. Publisher contact information may be obtained at . http//www. jstor. org/action/ human action inPu blisher? publisherCode=aea. . Each transcript of any part of a JSTOR transmission must contain the resembling copyright notice that appears on the screen or printed page of much(prenominal) transmission.JSTOR is a not-for- advantage service that helps scholars, researchers, and learners discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to maturation productivity and facilitate new forms of scholarship. For much information or so JSTOR, please contact emailprotected org. American Economic Association is collaborating with JSTOR to digitize, preserve and affix access to The American Economic Review. http//www. jstor. org CompetitionBetween Privateand Public Schools, Vouchers, and Peer-GroupEffectsBy DENNIS EPPLE AND RICHARD E. ROMANO* A suppositious and computational cast with tax income-financed, instruction- vacate customary domesticates and competitive, c are-financed mysterious rails is d eveloped. Students differ by expertness and income. Achievement depends on own booking and on adapteds abilities. Equilibrium has a strict creator structure of direct qualities and twodimensional educatee sorting with stratification by cogency and income. In unavowed inculcates, blue- readiness, low-income students receive learning discounts, fleck low dexterity, high-income students pay tuition premia.Tuition coupons plus the relative size of the closed-door domain and the extent of student sorting, and wellbeing high- faculty students relative to low- strength students. (JEL H42, 128) Discontent in the coupled countrys with the primary and secondary didacticsal system has become the norm. The worsen in SAT scores in the 1970s, embarrassinginternationalcomparisons of student act, slow festering in productivity mea sealeds, and alteration magnitude disparity in earnings tout ensemble call into question the bore of the preceptal system. Education poli cy figured prominently in recenit presidential elections.The grapple has centered on issues of trail pickaxe, including verifier systems (Kargonn De Witt, 1992). Typical voucher proposals yield students ensueing orphic domesticates a tax-financed, instruct-redeemable voucher of fixed issue forth toward (or perhaps covering) tuition. Although a 1993 California referendumfor vouchers was defeated, policy change at state and topical anesthetic levels abounds, as does change in the belowground educational sphither of influence. The state of manganese and educate districts in 30 states allow residents to choose the humans shoal their children run across. 2The city of Milwaukee introduced a voucher system in the 1989-1990 school year.A - bod f hush-hush o school and buck cloak-and-dagger- human race school initiatives be developing ( meet e. g. , joke F. Witte et al. , 1993 Steve Forbes, 1994 St level Glazerman and RobertH. Meyer, 1994 Joe Nathan, 1994 Newsweek, 1 994 Wall Street Journal, 1994 Steven Baker, 1995 Jay P. Green et al. , 1996). Educational reform emphasizing add school competition with an join ond * Epple GraduateSchool of IndustrialAdministration, Carnegie Mellon University, Pittsburgh, PA 15213 Romano Department of Economics, University of Florida, Gainesville, FL 32611.We greatly advise the comments of Linda Argote, Richard Arnott, Lawrence Kenny, Tracy Lewis, David Sappington, Suzanne Scotchmer, and three anonymous referees, in addition to workshop participants at Carnegie Mellon University, Florida State University, Indiana University, Northwestern University, Princeton University, the University of Chicago, the University of Colorado, the University of Florida, the University of Illinois, the University of Kansas, the University of Virginia, Yale University, the 1993 Public Choice meetings, and the 1994 American Economic Association meetings.We thank the National Science Foundation, and Romano thanks the Public Policy i nvestigate Center at the University of Florida for fiscal financial support. Epple ac agniseledges the supportof Northwestern University, where nigh(prenominal) of this research was conducted. Anv errors be ours. 2 Public funding of nonsecular schools and considerable warrantdom of school excerpt has been practicedfor years in England (Daphne earth-closetson, 1990) and much of Canada (Nick Kach and Kas Mazurek, 1986). These survival of the fittest systems support horizontal preeminence in learning and safeguards exist to enclosure upended ( property) specialisation.Our compend is interested primarily with the outlets of a voucher system on vertical differentiation. The provocatively titled report of the National Commission on righteousness in Education (1983), A Nation at Risk, details the decline of performance of U. S. students in the 1970s. More recent data can be found in Daniel M. Koretz ( 1987). Modest enlightens in performanceon standardized proceeding test s, followed by a leveling off, well below peak scores of the primaeval 1960s, characterizes the late 1980s and 1990s. 33 34 THE AMERICANECONOMICREVIEW role of the reclusive celestial sphere is at the forefront of he policy debate and recent policy initiatives. The modern-day economical shimmy for vouchers and augmentd educational choice was made by Milton Friedman (1962). The academic educational and political-science professions have since considered the pros and cons of voucher systems and educational choice (John E. Coons and Stephen D. Sugarman, 1978 Myron Liberman, 1989 John Chubb and Terry Moe, 1990). Economic synopsis of the interaction mingled with commonplace and mysterious schools, and of related policy instrumentslike vouchers, is only low to emerge. This paper continues the landing field of the merchandise for ducation by developing a lay that focuses on the interactionbetween the man and privy educational sectors and withal evidences the consequences of vouchers. We describe the compeerizer characteristics of the market for education with an open-enrollment frequent sector and a competitive hugger-mugger sector. Our proto eventful embodies two see elements of the educational process. First, students differ in their abilities. Higher office is as summateed to increase a students educational acquirement and that of peers in the school at run fored. Second, dwelling houses differ in their incomes, with higher(prenominal) income increasing the make for educational skill.A studentin our framework is consequently characterizedby an strength and a kinsperson income, a draw from a continuous bivariate distri entirelyion. A schools tint is determined by the destine ability of the student dead body, reflecting the models peer-groupeffect. We characterizethe proportionalitydistri exceptionof student faces across domain and esoteric schools and examine the tuition structure of cloak-and-dagger schools, expect that stud ent lineaments ar verifiable. We develop a theoretical and computational model in parallel, with the latter calibrated to existent judges of parameter pass judgments. Equilibria atomic compute 18 fictitious for a range of voucher assesss.Key characteristicsof an symmetryargon the following. A pecking orderof school qualities allow for be extradite, with the set of ( unvarying) frequent schools having the lowest-ability peer group and a strict ability-groupranking of surreptitious schools. The equilibrium student bodies of schools correspond to a partition of the ability-income-type seat of students with promenade 1998 stratification by income and, in many cases, stratificationby ability. As interpret 1 from our computational model illustrates, type space is therefore(prenominal) carved into aslope slices with each higher slice reservation up a hidden schools student body and with the bottom lice comprising the national sector. The normality of look at for a good peer group leads relatively high-income studentsto cross subsidize the schooling of relatively high-ability students, producing the latter partition. Private schools attract high-ability, low-income students by offering them tuition discounts, some measures fellowships. Even with free entry, schools toll discriminate by income against students who argon not on the margin between switching schools. The equilibrium differentiation of schools and economies of scale in education preclude perfect competition for any type of student.Nevertheless, this toll discriminationdoes not disrupt the internalization of the peer-group externality by private schools. An equilibrium without a public sector is P arto economical accustomed the equilibrium rate of schools. Because free public schools do not price the peer-group externality, an equilibriumwith public schools is P arto in debatent. In the computational model, we employ a Cobb-Douglas designate of benefit and educational ac complishment which incorporatesthe peer-group effect. The parameters are calibrated to U. S. data from various sources. We compute suppose equilibria for voucher alues ranging from $0 to $4,200 per student ($4,222 equals the outlay per student in public schools in 1988). With no vouchers, the predicted pct of students in the public sector is 90 component (the actual value for the United States is 88 percent). As the voucher is increased, the size of and mean(a) ability in the public sector decrease. With a $2,000 voucher, for example, the percentage of students remaining in the public sector equals 70 percent, and the mean ability declines by 15. 8 percent. 3 The integer exit of private schools in our model precludes existence of competitive equilibrium except in special cases.This integer puzzle and our approximation approachare talk abouted later in the paper. VOL. 88 NO. 1 EPPLE AND ROMANOPRIVATE-PUBLIC SCHOOLS opposition The entry of private schools and consequent to a greater extent expeditious sorting of students across schools caused by vouchers increases comely welfare (and achievement) only a little in our computational model, while having handsomer distributional set up. As we discuss in detail later, the magnitude of the aggregate effect depends on the extent of complementarity of peer ability and own ability in the educational production dish up. at that place is little empirical conclusion to accept assessment of the extent of uch complementarity. The voucher increases the premium to ability in private schools. The wallopingst proportionate gains from the voucher indeed accrue to low-income, high-ability students. For example, a base with income of $10,000 and student with ability at the 95th percentile has a welfare gain of about 7. 5 percent of income from a $2,000 voucher. Students of low income and low ability who remain in the public sector when a $2,000 dollar voucher is available experience low welfare losses but ba ffle up a majority. It bears emphasizing that our model takes public and private schools to be equally good providers of education, however.Some argue that private schools are more procreative and that the competitive effect of a voucher program will increase public-school effectiveness. 4 For example, Hoxby (1996) concludes from her empirical investigation that competition-induced performance improvement would increase public-school achievement by more than plenteous to offset 4 Caroline M. Hoxby (1994, 1996) provides evidence thatprivate-school competition increasespublic-school effectiveness. William N. Evans and Robert M. Schwab (1995) aim that Catholic private schools are more effective in inducing studentsto masterful high school and in addition to attend college.These studies take on the challenge of finding instrumentsthat predict well private-school care while being un cultivateal of unobserved determinants of educational achievement. Controversy exists concerning t he flavor of the instruments used. grab Thomas J. Kane ( 1996) for a discussion of Hoxbys methodology. David N. Figlio and Joe A. St unrivaled ( 1997) employ a different set of instrumentsthan Evans and Schwab and find that, at currentinput levels, religious private schools are less effective than public schools in producing achievement on standardizedexams in math and science but nonreligious private schools are more effective). discipline Witte (1996) and Figlio and Stone (1997) for references to another(prenominal) studies. 35 losses of the magnitude thiatemerge due to trim back peer property in our computational model. Our analysis delineates the allocative set up of vouchers and demonstrates a potential for significant redistribution. A theoretical-economics literature on education is beginning to emerge. Charles A. M. de Bartolome ( 1990) develops a twoneighborhoodmodel of the provision of public educational inputs ( woodland) with two ability types and peer-group e xternalities. He shows hat the voting/locational equilibrium is inefficient because the median voter does not internalize the consequences of migration on peer groups in choosing the input level. No independent income variability characterizes students in his model. Raquel Fernandez and Richard Rogerson (1996) introduce income differences in a two-neighborhood model of the provision of inputs but abstractfrom peergroup effects. They examine the effects of redistributive policies and direct controls on inputs. N all model has a private sector. Our analysis is place by its consideration of a private sector and its two-dimensional, ontinuous type space. In a nolrnativeanalysis of student groupings in the presence of peergroup effects, RichardArnott and John Rowse ( 1987) show how a societal plannerwould maximize the sum of achievements in allocating students of various abilities across classrooms. We analyze equilibrium outcomes, and most of our analysis is positive. Joseph E. Stigl itz ( 1974), Norman J. Ireland (1990), Ben Eden (1992), Charles F. Manski (1992), Michael Rothschild and Lawrence J. White (1995), Epple and Romano (1996), and GerhardGlomm and B. Ravikumar(1998) consider the consequences of a private sector for education. Stiglitz,Glomm and Ravikumnar, and Epple and Romano are concerned with the existence and properties of voting equilibria over taxfinanced, public-school expenditure in the presence of a private option. Ireland analyzes the effects of vouchers on utilities and the quality of the public alternative,taking the tax rate as exogenous. Individuals differ only by income, and the private alternative can be purchased continuously in all these analyses. Hence, the private sector is relatively passive, and issues of financial aid and differences in 36 frame 1998 THE AMERICANECONOMICREVIEW studentability across schools do not a rhytidectomy.Our model is distinguished by having differences in ability and related peer-group effects, and by pr oviding an active role for private-sector schools. Eden ( 1992) analyzes vouchers in a purely private market system of provision of education having two ability types and peergroup effects. A voucher equal to the difference between the social and private benefit of education to each ability type is shown to induce socially optimum provision of education. Key differences in our analysis include our consideration of the interaction between the public and private sectors, our geographic expedition of the implications of continuous ifferences in ability and income, and our anxiety to positive issues. Manski ( 1992) keep ups a computational analysis of vouchers that as well considers peer-group effects among other aspects of education (especially various intentions of public-school decision makers). Our models differ in a number of ways. Most importantly, we go for private schools to discriminate in their tuition policies, with many consequences. Rothschild and White ( 1995) analyz e a competitive model with consumers also inputs to production (a peer-group effect), employ higher education as their primary example.We share a concern for market determine in the presence of an externality. Differences in our model, among others, are the presence of a public sector, a more detailed specification of peer effects and carry for education, and student variation in both ability and household income. Our attention to the implications for price, profitability, and school qualities of a peer-group effect deriving from student abilities, the assignation of students correspond to ability and household income and the related distribution of educational benefits, and the effects of vouchers are not concerns in Rothschild and White.Private schools are cases of clubs with nonanonymous displace due to the abilitydependent externality and schools power to price it. Suzanne Scotchmer (1994) provides an excellent synthesis of this literature. We follow this literaturein our competitive specification of private schools as further discussed below. The next section presents the model. Section II develops the theoretical resultant roles. The computational results comprise Section III. Concluding remarks follow. An Appendix contains some of the detail. I. The get Household income is declared y, and each household has a student of ability b. The say arginal distributionof ability and income in the population is denoted f(b, y) and is assunmedto be continuous and positive on its . (0, bmax X (0, YmaxAll students support, S attend a school since we assume that free public schooling is takeredto no schooling. The household decision makers benefit escape, U( ), is increasing in numeraireconsumption and the educational achievement of the households student, and it is continuous and twice differentiable in both statements. Achievement, a = a (0, b), is a continuous and increasing use of goods and services of the students ability and the mean ability of t he student body in the school attended,O. Let Ytdenote after-taxincome and The influence of ability on own educational achievement is well documented and not controversial. Eric Hanushek( 1986) provides an excellent survey. In the economics literature, Anita A. Summers and Barbara L. Wolfe (1977) and Vernon Henderson et al. (1978) find significant peer-group effects. Evans et al. (1992) ad comely for excerpt bias in the formationof peer groups and show that it eliminates the significance of the peer group in explaining teenage pregnancy and displace out of school. They are careful to point out that their results should not be interpretedas suggesting that peer-groupeffects do not xist, but as demonstrating that scientific produce of those effects is inadequate. Note, too, that their work supports the notion that peer-group variables enter the returns function since a selection process does take place. The psychology literatureon peer-group effects in education also contains som e controversy. In their survey paper, Richard L. Moreland and John M. Levine (1992) conclude The fact that good students benefit from ability grouping, whereas poor students are harmedby it, suggests that the mean level of ability among classmates, as well as variability in their ability levels, could be an importantfactor.The results from several recent studies . . supportthis notion. This squares with our reading of the literature(Summers e and Wolfe, 1977 Henderson. t al. , 1978 Chen-Lin Kulik and James A. Kulik, 1982, 1984 Aage B. Sorensen, 1984 VOL. 88 NO. I EPPLE AND ROMANOPRIVATE-PUBLIC SCHOOLS ambition p tuition expenditure, the latter equal to cryptograph if a public school is attended. Thus, U = U(Y, p, a(O, b)), with U1, U2, a,, and a2 all positive. The achievement function prehends the peer-group effect in our model, discussed further below. To maintain simplicity and spotlight the role of peer groups, a chools quality is determined exclusively by the mean ability o f its peer group. 6In ongoing work we are extending the model to include variation in educational inputs. U is also fictive to recompense everywhere the single-crossing condition (SCI) (1) 0a OU/0 J yt ?. Preferences for school quality might also depend on ability. We say preferences satisfy run-down single crossing in ability if a Sorensen and MaureenT. Hallinan, 1986 Adam Gamoran and steel Berends, 1987 Jennie Oakes, 1987 Gamoran, 1992). However, in that location are alternative interpretations (Robert E. Slavin, 1987, 1990). For simplicity, the possibility that dispersion in peer ability also affects achievement is not built into our model. Roland Benabou (1996b) explores the consequences for economic growth of dispersion in human capital. 7 We suppose this to be uncontroversial. hile we k right forward W of no empirical studies that use direct measures of educational quality, a substantial empirical literature on the demand for educationalexpenditureexists. Although cons iderable smorgasbord in magnitudes of estimates of the income catch of demand for educational spending are present, estimates using a variety of approaches find the ncome ginger snap to be positive (Daniel Rubinfeld and Perry Shapiro, 1989). 8 Households may consider education a consumption good, an investment good, or a combination of the two. Our formulation can be interpretedto accommodate any of these pauperisms. However, for households not subject to borrowing constraints, a pure investment motive would imply a cipher income elasticity of demand. For such households, this in turn would imply that the SCI condition in (1) would be only flea-bitten satisfied. In light of the empir- OU/00 / Ob t OUIOyt which implies a weakly positive ability elasticity of demandfor quality.However, because the pertinent empirical evidence is merge and scarce, we postpone limiting preferences in this regard until necessary. 9 In our computational model and to illustrateour more oecumenical theoretical results, we acquire a Cobb-Douglas specification of the gain function (2) Hence, for students of the identical ability, any indifference curve in the (0, p)-plane of a higher-income household cuts any indifference curve of a lower-income household from below. This condition corresponds to an income elasticity of demand for educational quality that is positive at all qualities for all types. One set of sufficient conditions on U for SCI is U11 0 and U12 2 0, with at least I one having strict inequality. 8 37 U = (yt-p)a(O, a(O, b) b) = 0Yb6 g O y O. While(2) satisfiesSCI,it embodies he neut tral assumption of vigor ability elasticity of demand at O? /0 9b 0. Our computational results are not driven by own-ability effects on the demand for education. Keep in mind, too, that the theoretical results do not assume specification (2). A schools salute depend only on the number of students it enrolls, since inputs part only with size. All schools, public and private,hav e the simple equal function (3)C(k) = V(k) + F V 0 Vo ical evidence suggesting the income elasticity to be positive, we conserve space in the development that follows by anticipate that SCI is strict for all households. 9 Henderson et al. (1978) find no interaction between own ability and the benefits to an meliorate peer group, corresponding to 2IU/00&b= 0 in our model. Summers and Wolfe ( 1977) find some supportfor higherpeer-group benefits to lower-ability students, that is, 02U/la6ab 0. Thus the literatureprovides limited evidence from which to draw conclusions. 38 THE AMERICANECONOMICREVIEW where k denotes the number of attendancestudents.Technical differencesamong schools are not an element of our model (for simplicity). Hence, vouchers cannot drive technically inefficient schools from the market,an effect predicted by some proponents of vouchers (see footnote4). Let k* denotethe efficientscale, (4) k* ARGMINC(k)/k. The presumptionof some economies of scale in education i s realistic (Lawrence Kenny, 1982) and important. Otherwise, the private market would stir an unnumberable number of schools containing infinitely refined peer groups. Our models equilibriumwill be conformable with the fact thatthe numberof types of studentsgreatlyexceeds the numberof schools.Public-sector schools offer free opening to all students. This open-enrollment policy leads to consistent public schools in equilibrium because we assume no frictions in public-school choice are present. Without equalization of 0s in public-sector schools, students would migrate to higher-0 schools to reap the benefits of a punter peer group. With equalized 0s, no incentives for switching schools within the public sector remain. We study the alternative of neighborhood school systems that impose residence enquirementsin Epple and Romano (1995). Since all public schools will have the same , one can think of the public sector as consisting of one (possibly large) school. Publicsector schoo ling is financed by a comparative income tax, t, paid by all households, whether or not the households child attends school in the public sector. Thus, Yt= (1 t)y. The public sector chooses the (integer) number of schools and their sizes to minimize the rack up be of providing schooling subjectto (3). The tax rate ad adeptsto balance the cypher. Because households are atomistic, there is no tax consequence to a households decision about school attendance. The public finance of chooling can thusly be largely suppressedin the analysis until the consideration of vouchers. The public sector is passive in this model for simplicity. Public-sector schools do not segment students by ability (track), increase educational inputs to compete more effectively with the private sector, or behave strategically sue 1998 in any way. More realistic alternativesare importanttopics for research, some of which are discussed in the final section. Private-sectorschools maximize cabbage, and there i s free entry anidexit. 10Modeling private schools as choosing an gateway policy and uitionpolicy is convenient andinvolves no loss of generality. Student types are observable, implying that tuition and admission can be conditioned on ability and income as competition permits. 1 Private schools are an example of clubs with non-anonymous crowding (Scotchmer and Myrna H. Wooders, 1987 Scotchmer, 1997) because of the peer-groupeffect, and we model private-schoolbehaviorfollowing the literature on competitive club economies. In particular,private schools maximize benefit as utility tak-ers(see Scotchmer, 1994), a generalization of price-taking when consumers (types) and productsdiffer. Private chools believe they can attract any studenttype by offering admission at a tuition yielding at least the maximum utility the student could obtain elsewhere. Let an i subscript, i = 1, 2, .. n, indicate a value for the ith private school. A slide fastener subscript does the same for the public sc hool. Let pi (b, y) denote the tuition necessary to enter school i, with po(b, y) = 0 V (b, y). Let ai (b, y) C 0, 1 denote the proportionof type (b, y) in the population that school i admits, 10Consideration of alternative objective functions to profitmaximization is reasonable,especially effrontery the significant proportion of nonprofit schools.Some private pursuethe objective of quality schools might, for exa-mnple, maximization. forest maximization, like profit maximization, is a member of a set of objective functions that are utility independent in the sense that they place no weight on offering any student types higher utility than the students (equilibrium) reservation utility. Our preliminary analysis of this issue suggests that equilibtia where some private schools pursue objectives from this set other than profit maximization must also be competitive equilibria. Roughly, the failure of any school to maximize profits would permit ently by a profit-maximizingschool. The n otion is that abilities can be determinedthrough testing, and required financial disclosures permit determination of household income. At least in the case of Cobb-Douglas utility, equation (2), students will have no incentive to underperformon exams, since tuition will be nonincreasing in ability in equilibrium (proved in Epple and Romano 1993). bonus compatibility in the reporting of income is more complex. EPPLE AND ROMANOPRIVATE-PUBLIC SCHOOLS COMPETITION VOL. 88 NO. I with any ao(b, y) E 0, 1 optimal for the public school as determinedby the residualdemand for public education.A private schools profit-maximizationproblem can be written as (5) MAX rri Oj,kj,pj(b,y),aj(b,y) pi(b, y)ai (b, y) f s X f(b, y) db dy-V(ki) F subject to ai (b, y) E 0, 1 V (b, y) (5a) (5b) (b, y)a(Oi, U(y,-pi MAX 2 j b) ) p1(b, y), a(0j, b)) j * i aj(b,y) O is in the optimal set of j U(yt I0,1, ,n that (5b) hold for all (b, y) as we have specified (i. e. , including for nonadmitted students). Tuition charged to students for whom ai (b, y) = 0 is school is only optimal choice (i. e. , nonadmittedstudents) is irrelevant. Note, too, that tuition such that (5b) holds with strict quality will be optimal. Private schools enter so long as they expect to make positive profits as utility takers. Because officer private schools maximize profitsas utility takers,entryresults if and only if wri 0 for some officer school. The public-sector/private-sectorequilibriumis described by the following five conditions in addition to the government activity balanced-budget condition presented below in Section II, subsection C, for the more general case with vouchers. Condition UM U*(b, y) MAX U(y,-pi(b, V (b, y) 39 ie E 0,1, ,nI ai(b,y) y), a(0i, b)) O is in the optimal set of iV (b, y). (5c) (b, y)f(b, y) db dy ki =fai Condition VIM s Oi, ki, pi(b, y), ai (b, y) satisfy (5), (Sd) O kjfbai(b, y)f(b, y) db dy. s Constraints(5c) and (5d) define, respectively, the size of the schools student body a nd the mean ability. Constraint (5a) precludes a school from admitting a negative number of a type or more of a type than exits in the population. 2Constraint(5b) imposes the utilitytaking assumption. Students alternatives are limited to schools where they are admitted. Students always have the option of attending the public school. It is innocuous to require i= 1,2, ,n.Condition ZfH 7ri = 0 i = 1, 2, , n. Conditions PSP po(b, y) = V (b, y) ao(b, y) E 0, 1 V (b, y) 12 One might object to the presumptionthat competitive schools recognize the limit to demand. The presumption is analogous to a monopolistically competitive firms light of a limit on its demand curve. Dropping the presumptionwould lead to schools admitting infinite densities of some types. See Scotchmer (1994) for the analogue in the literatureon club goods. ao(b, y)f (b, y) db dy ko= s Go =-Af ko bao(b, y)f (b, y) db dy. s THE AMERICANECONOMICREVIEW 40 Condition MC n xai (b,y)=1 V(b,y). i=OCondition UM summarizes house hold utility maximization. Households choose a mostpreferred private or public school, taking admission/tuition policies, school qualities, and taxes as given. Profit maximization of private schools (VIM) and the public-sector policies (PSP) have been discussed. While the entry assumption higher up is full-dressly part of the explanation of equilibrium, it is convenient to substitute the implication that private schools must earn zero profits (ZH). The last condition is market clearance, which uses the simplifying assumption above that free public schooling is preferredto no schooling.II. Theoretical esults R A. Solution to the Private Schools Problem Using UM, the runner-order conditions for problem (5) can be written as follows U(yt- pi , a(Oi, b)) (6a) =U*(b, ai (b, y) (6b) piC as f (6c) We now turn to the properties of equilibrium, anticipate one exists. Existence issues are discussed below. Heuristic argumentshave been substituted for formal proofs when reasonable. The fir st result concerns the qualities of schools. b0,db (b, si) L sho0 V(ki) yd 0 +io (Oi b) = n7i- X PROPOSITION 1 A strict hierarchy of school qualities results, with the public sector V (b y) i i J ith equality combined with the equilibrium condition UM pe () is student-type (b, y)s reservation price for attending school of quality 0i. Condition (6b) characterizes optimal admission policies. The term 77i0i b) may ( be thoughtof as the peripheral appeal of admission run via the peer-group externality in school i. From (6c), 77ithe Lagrangianmultiplier on (5d) equals the per-studentrevenue change in school i deriving from a change in 0i. The appropriatelyscaled change in 0i due to admitting student of ability b equals (b t 0k) its negative is therefore multiplied by rqj o obtain the peer-externalitycost.The peer cost of admitting students with ability below the schools mean is positive because the resulting quality decline dictates bring down tuition to all students, while the peer cost of admitting above-mean-ability students is negative. Let ( MCi (b) V(ki ) + r7i 0 b), which we term effective peripheral cost. Types with reservation prices below MCi (b) are not impulsive to pay enough to cover their effective peripheral cost and are not admitted. The school admits all of a type that has a reservationprice above effective marginal cost, and any ai E 0, 1 is optimal if pi* = MCi. 1 B. Properties of EquilibriumV (b, y) y) MARCH 1998 ai(b y) f(b y) db dy Condition ( 6a) describes sclhooli s optimnalut ition function, Pi* (b, y, Oi) and is sound (5b) 3 Results (6b) and (6c) are found by substituting p* from (6a) into (5), and then forming a Lagrangianfunction to take account of (Sc) aind(Sd). Result (6b) is then derived by pointwise optimisation over ai while taking account of the constraint (Sa). 4 In the upper and lower lines of (6b), the solution for ai is at a corner, and the first-orderconditions are also sufficient for a local maximum. In the ni tty-gritty line of (6b), where p * MCI and any ac (b, y) E 0, 11 satisfies the irst-orderconditions, V sufficiently large implies local maximization. VOL. 88 NO. 1 EPPLE AND ROMANOPRIVATE-PUBLIC SCHOOLS COMPETITION having the lowest-ability peer group 0,, fJn-I .. fJI 00. Formal proof is in the Appendix. here an economic interpretationis provided. All private schools must be of higher peer quality than schools in the public sector. Otherwise, no students would be willing to pay to attend any private school. Why must a strict hierarchy of private schools characterize equilibrium? If two private schools were of the same quality, then they would compete perfectly for students.Consequently, they would have the same effective marginal be of admitting all types, and their tuitions (to all admitted students) would equal effective marginal costs. An opportunity to increase profits would exist by varying admissions/tuitions in such a way to either (a) increase quality and admit a stude nt body that values quality by more, or (b) decrease quality and admit a student body that values quality by less. In either case, the school differentiates itself in quality, at the same time attractinga student body that permits profitable price discrimination over the quality change.We sketch the example of a profitablequality improvement, beginning with schools having identical student bodies (the proof shows that this is without loss of generality). Let one school admit the same numberof (b2, Y2) types as it expels of (bl, yl) types, where b2 b, and Y2 Yl, implying an increase in 0 but no change in production costs, V(k) + F. Further, choose the types (which is always feasible) such that Y2 YI b2 bi by enough that, using SCI, the (b2, Y2)-types value increased quality by more than the (bl, Yi)types, even though their abilities differ.This permits the school to charge the newly admitted students tuitions higher than their effective marginal costs because they are selected to value quality increases by more than the expelled students. The profit increase occurs because the new student body values the quality increase by more than would the original student body 0 and rl rise in the school. It would not increase profits to substitute students in such a way that 0 rises without also changing the student bodys norm value of quality improvements, because tuitions equal 41 ffective marginal costs in the initially nondifferentiated schools. 5 This example assumes that a school substitutes students to increase quality, but alternativeprofitablesubstitutions exist that decrease quality, roughly, by also creating a lower-income studentbody. In either case, the argumentdepends on SCI. It also identifies the models force for diagonal stratification (see the examples in Figure 1). As developed more to the full below, this stratification results because students having relatively high income and low ability within a school cross subsidize relatively lowincome, hig h-ability students.The strict hierarchy of proposal 1 supports the equity-relatedconcerns of some that private schools operate to the detriment of public schools by siphoning off higher-ability students. Whethera strict hierarchyis efficient is analyzed below. First we develop furtherthe positive properties of equilibrium. offer 2 describes equilibriumpricing, and offer 3 describes the resulting partition of types. Some definitions are useful. Let (b, y) E SIai (b, y) 0 is optimal denote the admission space of school i, i = 0, 1, , n (see Figure 1, for example). A locus of points (b,y) E A. n Aj, i j, assuming t exists, is referredto as a limitation locus between i and. (Boundary loci have zero measure in S, as proved in Epple and Romano 1993. ) Since any household prefers free public schooling to no schooling, the entire type space S is partitioned into admission spaces. Last, to avoid tedious qualification of statements for public-sector schools, we specify that MCo -0 for a ll (b, y). This notation is convenient since students see a zero cost of public education. PROPOSITION 2 (i) On a limit locus between school i and j, pi MCi(b) and pj = MCj(b) pricing on bourne loci is strictly according to ability in private schools. ii) pi (b, y) MCi (b) for off- limit pointstudents who attend private school i pricing off- Mathematically,beginning with equal Os,first-order effects on profits of varying admissions vanish, but the profit function is broken-backed in some directions in a(b, y), p(b, y) -space, allowing a profit increase. 42 THE AMERICANECONOMICREVIEW boundary loci depends on income in private schools. (iii) Every student attends a school that would maximize utility if all schools instead set pi equal to equilibrium MCi Jor all students. The allocation is as though effective marginal cost pricing prevails in private schools. 16See Epple and Romano (1993) for proof. Competition between private schools that share a boundary locus forces prices to e ffective marginalcosts for student-typeson the locus. These students are absent-mindedto attending the schools sharing the locus. Private schools then have no power to price discriminatewith respect to income on boundaryloci. Prices are, however, adjusted to differing abilities because private schools internalize the peergroup effect. Tuition to private school i decreases with ability at rate rRialong its boundary loci, reflecting the value of peergroup improvements of the schools student body.Moving inside a boundarylocus in a private schools admission space, students preferences change in such a way that they would strictly prefer the school attended if it practiced effective marginal-cost pricing. Part (ii) of Proposition 2 establishes that private schools exploit this by increasing price. These students are also indifferent between the private school attended or their best alternative by (6a), but this is a result of discriminatory pricing. Generally, then, price depends both o n ability and income within admission spaces. 7 Part (iii) of Proposition 2 follows because it is profitable for a private school to be sure o attractany student whose reservation price 16 The statementsregard the equilibriumeffective marginal cost. Income effects would cause these costs to change if tuition equaled effective marginal cost for all students. This has distributional (but not talent) implications. 7 While there are no published studies of the allocation of financial aid by income and ability among private elementary and secondary schools, there is evidence on the allocation of financial aid by colleges and universities. There the evidence is that both ability and family income are significant determinantsof whether and how much financial aid is received (J.Brad Schwartz, 1986 Sandra R. Baum and capital of Minnesota Schwartz, 1988 Charles T. Clotfelter, 1991). MARCH 1998 exceeds the schools effective marginal cost. The student allocations link to effective marginal cos ts, and hence abilities, will be shown to be efficient (except for the public sector). The income-related price discrimination that occurs does not disrupt the allocation consistent with effective marginal-cost pricing rather,it is purely redistributive. While this income-related price discrimination is of the first degree (a la Pigou), its magnitude is limited by competition for students among the differentiatedschools.Near a boundaryin a schools admission space, a students preference for the school attended would be slight under effective marginal-cost pricing, so that the admitting school can capture little rent. The numberand sizes of private schools then determine their powier to price discriminate over income. All private schools have student bodies less than k* by a quasi(prenominal) argumentto that in more standardmonopolistically competitive equilibria. 8Here school is marginal-revenuecurve can be constructed by ordering from highest to lowest students reservation prices m inus peer costs i. e. , p* + rbi(b Of), and thus the associated ownward-slopingaverage revenue curve may be derived. Zero profits then implies a scale below k*. If we let k* decline, then private schools become more numerous and less differentiated (have closer 0s), and incomerelated price discriminationdeclines. Now consider the partition of types into schools. We say stratificationby income (SBI) holds if, for any two households having students of the same ability, one households choice of a higher-O school implies it has a weakly higher income than the other household. Analogously, stratification by ability (SBA) is present if, holding income fixed, the household that chooses a higher-Oschool must ave a student of weakly higher ability. The combination of SBI and SBA implies a diagonalized partitionas, for example, in Figure 1. PROPOSITION 3 (i) SBI characterizes equilibrium. (ii) If preferences satisfy weak c single crossing in ability (W-SCB) and m7, 18 The points made here ar e proved in Epple and Romano (1993). EPPLE AND ROMANOPRIVATE-PUBLIC SCHOOLS COMPETITION VOL. 88 NO. I ?72 ? - ? 71, then SBA also characterizes equilibrium. 9 To confirm part (i), consider two households with students of the same ability but dif- feringincomes y2 yI. In the (0, p) -plane, indifference curves of a household are upward loping. For the same ability, SCI implies that any indifference curve of y2 cuts any indifference curve of yl from below. Allocations are as if tuitions equal effective marginal costs part (iii) of Proposition 2. Thus, the choice between schools i andj may be representedin the ( 0, p ) -plane as a choice between ( Oi, MCi (b)) and (0j, MCj(b)). If Oj Oi, it must be that MCj(b) MCi (b) if either type chooses i. A standard single-crossing argument then applies to complete the proof. Part (ii) is proved in the Appendix here we provide some intuition. Assume first that the demand for quality is independent of ability (e. . , as in the Cobb-Douglas specifi cation) and that all private schools give the same discount to ability along their boundary loci (i. e. , schools 7s are the same). Holding nominal household income fixed, real income would rise with student ability due to tuition discounts at all private schools. SBA would then result by the same logic explaining SBI. Hence, the combination of a positive income elasticity (SCI) and discounts to ability alone would cause both SBI and SBA, the diagonalized partition as in Figure 1. Relatively high-income and lowability students cross subsidize relatively low-income and high-ability students in rivate schools. The argument holds more strongly if the 7s strictly ascend or if WSCB holds strictly. However, neither condition is necessary for SBA, nor do any of our other results require these conditions or SBA. It may be possible absent these conditions to get cases having nonmonotonic boundary loci in the (b, y) -plane. 20 9 We thank an anonymous referee for encouraging us to investigate bid-rentfunctions (see e. g. , MasahisaFujita, 1989), which ultimately led to part (ii) of Proposition 3. 20 The alternativeto W-SCB implies that lower-ability types are willing to pay more for a better peer group, and he alternativeto weakly ascend qs implies that lower- 43 We now turnto normativeresults which are quite intuitive. Again, see Epple and Romano (1993) for the formal analysis. Pareto efficiency requires (i) a student allocation that internalizes the peer-group externality given the nmtmberf schools, and (ii) entry as long o as aggregate household net willingness to pay for an allocation with one more school exceeds the change in all schools costs. An equilibrium without a public sector would satisfy condition (i) but not condition (ii). Effective marginal cost includes the marginal value of he peer group externality, implying that MCi (b) equals the social marginal cost of attendance at school i by a student of ability b. A purely private-school equilibrium then satis fies efficiency condition (i) by part (iii) of Proposition 2. However, entry to the point of zero profits entails externalities so that efficient entry condition (ii) fails to hold in a full private equilibrium. An entrantcapturesthe full value of its product to the studentbody it admits but ignores utility changes of nonadmitted students and profit changes of other schools resulting from the reallocation. Fixed costs, quality schools give bigger discounts to ability. all would tend to work against pure ability stratification, though Proposition I implies that some degree of ability stratificationwould be present. It is desirable to demonstrate SBA without assuming ascending qs, since these values are endogenous. However, providing general primitive conditions for SBA independentof assumptionsconcerning the equilibrium qs is difficult, because their equilibrium values depend on the entire distribution of types in the population. For the Cobb-Douglas case and ssuming independence o f income and ability in the population, we (Epple and Romano, 1993) have shown SBA without assuming weakly ascending 7s. 21 The comparison of the equilibrium number of schools in a fully private equilibrium to the Paretoefficient number entails a trade-off. The entrant ignores the lost revenues and cost savings to other schools from the students that t admits. Since almost every student ati tracted away from incumbent schools is inframarginal (i. e. , tuition exceeds effective marginal cost), the net effect here of entry is negative, tending to cause too much entry.Opposing this is the entrants failure to capture the full returnsfrom increased varie-tyof school qualities that results. Altlhoughthe entrant fully price discriminates to the students it admits, it cannot tax other students for the adjustments in the incumbent schools qualities. A net benefit to other students is likely to result because the incumbent schools will better accommodate preferences. 44 THE AMERICANECONOMICRE VIEW hence the finite size of an entrant,underlie the entry externalities as in many models of monopolistic competition. Introductionof the free public sector implies eviations from both efficiency conditions. In general, the public sector displaces multiple differentiatedprivate schools, substitutingthe equivalent of one large homogeneous school. This effective reduction in the number of schools is without attention to costs and benefits, principally implying a deviation from efficiency condition (ii). Holding fixed the number of schools in the public-private equilibrium (and counting the public sector as one school), zero pricing of public schooling violates condition (i). By just reallocating students between the public sector and private school 1 uprise their shared oundary locus, Paretian gains are feasible. Reference to the upper panel of Figure 1 from our computational equilibrium may help clarify the argument. Gains would result from slip into private school I relatively lowerability students below but pricey the boundary locus, students for whom the marginal social cost in the public sector is positive. These students are nearly indifferent between the two schools when facing the social cost of attending the private school but a tuition (zero) below the social cost of attending the public school. Students near the boundary locus and ttending the private school may also be of sufficiently high ability that the social cost of attending the public school is negative. Gains from shifting such studentsinto the public sector are then also feasible. Such students exist in our computational model, the rough prescriptionbeing to rotatethe boundarylocus counterclockwise at the point of ability having zero social marginal cost in the public school. Collecting these results, we have the following proposition. efficient. (ii) The public-private-sector equilibrium has neither an efficient number of schools, nor an efficient student allocation iven the number of schools. When fixed costs of schooling are small, the departure from efficiency in a fully private equilibrium will be correspondingly small. Part (i) of Proposition 4 can then be interpreted as making a case for private schooling and the vouchers we study. However, we have some reservations concerning this efficiency result. First, we are sympathetic to the view of many that access to a quality education is a right and serves as a means to limit historical inequities. Second, longer-run externalities from education not considered by private schools, like reduced crime, may be present.For these reasons, we explore the consequences of vouchers on all types instead of just providing aggregate measures. A somewhat distinct concern arises because exact equilibrium exists only in special cases. The interpretationof the efficiency results in the approximateequilibriumwe study is discussed in subsection D, below. C. Vouchers We examine tax-financedcash awardsto all those attending private school. 22 No role for vouchers is present in the tuition-free public t sector. Refo-rmulate he model by everywhere adding the amount of the voucher, v, to yt for households that choose a private school.The governments budget constraintis tyf (b, y) db dy (7) s This positive externality will tend to cause too little entry. We believe that too many or too few private schools are possible, but we have not proved this. vf(b,y)dbdy fJ PROPOSITION 4 (i) The allocation in a fuilly private equilibrium is (Pareto) efficient given the number of schools the equilibrium number of schools is not, however, more often than not MARCH 1998 . U AlU UA2U . +srs N U . +r , / B 7(k 22 Our model permits households to retain as income any excess of the voucher amount over the tuition paid to he private school of choice, thereby avoiding considerable complication. VOL. 88 NO. 1 45 EPPLE AND ROMANOPRIVATE-PUBLIC SCHOOLS COMPETITION where N and k denote, respectively, the costminimizing number and size o f schools in the public sector that satisfy demandfor public education. Vouchers lower the real price of private education and increase the demand for it. We examine the effects of vouchers in our computationalmodel. the results will provide at least suggestive evidence about the impact of policy interventions. However, scrimp empirical evidence exists on some important parameters of the odel. D. Existence of Equilibriumand an ApproximateEquilibrium We require specifications for the density of income and ability, the utility and achievement functions, and the cost function for education. As we discuss in the Appendix, exact equilibrium generally fails to exist due to the integer number of private schools. We examine an approximate equilibrium in our computational analysis. Our epsilon-competitive equilibrium requires that no (utility-taking) private school, incumbentor entrant,could increase profitsby more than s. Let 7rax and .. min denote the maximum and lower limit profits arned by incumbent schools which maximize profits overp(b, y) and a (b, y) locally, and replace Zfl in the definitionof equilibrium with MAX irmax, Xrmax -lrmin rminl c Here lrmax equals the maximum potential profits to an entrant,and the maximum of the second two ground in the brackets equals the largest feasible profit increase by an incumbent school. The rewrite definition of equilibrium continues to require UM, PSP, MC, and local profit maximization by incumbent private schools i. e. , (6a) (6c). Last, the number of private schools is the stripped-down number material these requirements.The epsilon equilibriumretains all the positive propertiesof an exact equilibriumexcept that private schools could gain s in profits via global adjustments. The allocation of students in a fully private equilibriumwould then continue to satisfy efficiency condition (i). M III. computational quilibrium odeland E R Illustrative esults We develop a computationalmodel to illustrate our results, to e xamine vouchers, and to explore issues for which comparative-static analysis may yield ambiguous results. We calibrate it to existing empirical evidence so that A. Specification and CalibrationWe assume that n (b) is distributedbivar- iate normal with mean LbJand covariance matrix 2 01b P UbUy P bUy aY J To calibratethe distributionof income, we use mean ($36,250) and median ($28,906) income for households from U. S. census datafor 1989. With units of income in thousands of dollars, these imply that ,uy = 3. 36 and ay = 0. 68. We adopt specification (2) for the combined utility-achievement function. To calibrate the ability distribution we presume that educational achievement determines futureearnings and that the benchmarkeconomy is in a steady state. First, define normed achievement, aN, s our achievement function raised to the power 1/3 and multiplied by a constant, aN Y Ka = KO lb. 23 Then, a studentwith ability b attending a school with a peer quality of 0 is presumedto have futureannualearnings (E) given by ln E = ln aN= In K + (y/o/)ln 0 + ln b. This normalization is such that a percentage change in ability leads to the same percentage change in dollars earned. Henderson et al. ( 1978) reportthe change in achievement percentile that results from moving students from classes stratified by ability to mixed 23 The constant of proportionality, K, is arbitrary. A onvenient scaling is to set K = E b -. This scaling has the propertythat, if all students in the populationwere to attend the same school (i. e. , 0 = Eb), then normed achievement would equal ability (i. e. , aN = b). 46 THE AMERICANECONOMICREVIEW classes. An elasticity of achievement with respect to peer ability that is 30 percent as large as the elasticity with respect to own ability is representative of the results they report. We adopt the somewhat conservativevalue of y/l 0. 2. To complete the calibration of the distribution of ability, we then assume that the observed household-income distri butionis the ncome distribution that emerges in a steadystate equilibrium in our benchmark model. 24 This yields Ilb = 2. 42 and b = 0. 61. Thus, mean and median ability are 13. 6 and 11. 3, respectively, and the standard deviation of ability is 9. 1. 25 GarySolon (1992) and David J. Zimmerman (1992) provide evidence on the correlationbea tween fathers income and sons incomrre,nd they both find that the best point estimate of this correlationis approximrately. 4. Intergen0 erationalcorrelationin income arises from two sourcescorrelationbetween householdincome nd student ability and, for given ability, correlation between income and quality of school attended. Hence, SBI suggests that the intergenerationalcorrelationin incomes is an upper bound on the correlationbetween parents income and childs ability. For purposes of sensitivity analysis, we then assume that p E 0, 0. 4. For our benchmarkcase, we set p = 0, More precisely, we let the distribution of ability be lognormal, and we ap proximateby assuming that this generates a lognormal distributionof earnings. We set the first two moments of the distribution of earnings equal to the irst two moments of the distribution of income. That is, a we choose 11h nd cb such that our benchmarkequilibrium has EaN = Eylm and VaraN = Vary/m2. The constantm is the ratio of employed workersper household to the number of students per household (m = 2. 6 in 1990). The distribution of earnings will not be exactly lognormal because of the discrete difference in schools attended, even though the distribution of ability is presumed to be lognormal. If every student attended public school in the benchmarkmodel, and hence faced the same 0, earnings would be exactly lognormal.The approximation is a good one because 90 percent of the students do attend public schools as we will see. 25 cogency can be related to IQ. Using IQ X( cytosine, 256), one obtains In b = -1. 38 + 0. 038(IQ). In our novoucher steady state, this implies that a wo rkerwith an IQ of 100 has expected income of $22,074, and a 10-point increase in his IQ increases expected income to $32,510. See the discussion in what follows relatingto Figure 6 and the calculation of expected steady-state income conditional on ability. 24 MARCH 1998 which is particularly onvenient for our steadyc state calibrationof the model.This completes the calibrationof f(b, y). We now complete the calibration of preferences. The Cobb-Douglas specification implies unitary price and income elasticities for school quality, 0. Given the absence of empirical evidence on the demand for quality, these are plausible central values and are consistent with estimates of demand for school expenditure (see e. g. , Theodore Bergstrom et al. , 1982). This function also implies thatthe marginal rate of substitutionbetween school quality and the numeraire is invariant to own ability. Empirical evidence is mixed about whether an improvementin peer group is more eneficial to high- or low-a bility students. Hence, our models assumption that the effect of peer group is not prejudice toward either highor low-ability types seems an appropriate choice for a baseline model. If school quality could be purchasedat a constant price per unit of quality, each households expenditure on education relative to total expenditureon other goods would be y/( 1 + -y). The existing share of aggregate disposable personal income in the United States that is spent on education is most 0. 056. Hence, we set y = 0. 06. Using y/P = 0. 2 from above, the calibrated tility-achievement function is then U = (Yt P)0006b0. 30. We chose a cost function that is quadratic in the percentage of students (or households) a school serves F + V(k) = 12 + 1,300k + 13,333k2, with parameters set as follows. intake per student in public schools in 1988 was $4,222 (Statistical Abstract, 1991 p. 434) and there was 1/2studentper household (Statistical Abstract, 1992 pp. 46, 139). We specified our benchmark case t o have four private schools and chose parametervalues such that average cost in equilibriumwas approximately$4,200 per pupil. 26 Experimentation indicated that 6 We have presented the cost function in terms of the percentage of students served or, equivalently, the per- VOL. 88 NO. I EPPLE AND ROMANOPRIVATE-PUBLIC SCHOOLS COMPETITION equilibriumpropertiesare not very sensitive to the benchmark number of schools, but rather are sensitive to the tokenish of the average cost of schooling. We set e = 4. 2. This is the minimum value sufficient to pick up existence of epsilon equilibrium for voucher values varying from zero to $4,200 per student. 27 B. Results For our benchmark equilibrium with no voucher, the public sector has 90 percent of the student population, and the four private chools combined serve the remainder. The actual U. S. percentage of students enrolled in public schools during this period equaled 88 percent. Increasing p from 0 to 0. 4 reduces public-sector attendance to 88 percent. Effects on other variables of so changing p are also small, and the results that follow are for p = 0. centage of households served, k. In terms of k, average cost reaches a minimum at $2,100, with k* = 0. 03 $2,100 can then be interpretedas the average cost per household. There are twice as many households as students in the United States. Letting s denote the numberof studentsand ubstituting s = k/2, one sees that the minimum of the average cost per student is $4,200. In our presentation,we focus on per-student measures of tuition and costs the related per-household measures are simply half those of the per-studentvalues. 27 This value is about 7 percent of the cost of a school operational at a scale that minimizes cost per student. Relative to fixed cost, ? is approximately 35 percent. Of course, a minimal E, however measured, is desirable. We have studied how the minimum e varies as we vary efficient school scale, k*, while holding average cost constant.We find t hat the requisite e to support equilibrium varies approximatelyproportionatelywith k* if fixed cost is varied proportionately with k*. This suggests, as we would expect, that e can be made as small as desired if k * is made sufficiently small. We have also investigated increasing fixed cost while holding k* and minimum average cost constant. This tends to reduce the ratio of e to fixed cost but increases the absolute magnitude of e required to sustain equilibrium. Our investigation reveals that substantive findings from the computational model re not sensitive to the choice of k* or the relative magnitude of fixed to variable cost. Rather, the key aspect of costs is the value of average cost at the minimum, and as discussed in the text, this value is based on observed school costs. The problem with pursuinga calibrationthat further lowers e is that it leads to a computationally unmanageable number of schools for large vouchers. 47 Othercomputationalresults are presentedin Figures 1- 6. The upperpanel of Figure 1 presents the boundary loci and admission sets in type space, in addition to the equilibrium Os nd ks. Here and in some other figures, both absolute and percentile ability scales are provided for perspective. The lower panel displays the allocation for a voucher of $1,800. The linear boundary loci derive from the Cobb-Douglas specification. For results we present, intersections of boundaryloci, if any, occur very near the bounds of the support of type space. S