The Pros and Cons of Free College for All Discussion

The Pros and Cons of Free College for All Discussion

ORDER A PLAGIARISM FREE PAPER NOW

8 Chapter 8 in Economic Forces at Work: Selected Works by Armen A. Alchian, Libery Press, Inianapolis, 1977 The Economic and Social Impact of Free Tuition arely do educational issues provoke as much passion as the proposal to raise tuition fees in California colleges. Unfortunately, the passion has not been matched by reason-it is hard to find a clear statement of the consequences of or reasons for a zero tuition or a high tuition fee. It is hard to determine from the public comments whether the antagonists differ about what the consequences of alternative tuition arrangements would be or have different preferences with respect to well perceived consequences. Some defenders of zero tuition have asserted that zero tuition is necessary for aid to poorer students, for the maintenance of our great system of higher education, for the preservation of free and prosperous society, for achievement of great social benefits, for educational opportunity for all, is a hallowed century-old tradition, and that tuition is a tax on education. Some proponents of tuition fees have argued, for example, that the university and colleges are harboring delinquents who R Acknowledgment is made to the Lilly Endowment, Inc., for a research grant to UCLA during which the present article was written. The opinions expressed here in no way reflect any conditions of that research grant. 204• Economic Forces at Work would not be there with full tuition, the poor are aiding the rich, students should pay tuition in order to appreciate their education, taxes are excessive, and low tuition requires exploitation of an underpaid faculty, to cite a few. Most of these arguments are so patently fallacious or nonsensical or irrelevant that they do disservice to the more intelligent arguments. But there are some propositions that merit closer examination. To evaluate them it is first necessary to identify at some length the issues that are involved in analyzing and thereby choosing among the alternatives-and in the process make clear my own preferences. If I overlook significant objectives or consequences, perhaps others will be stimulated to fill the gaps. The issues represent a classic topic for applied economics-the effects of different means of allocating scarce resources !}mong competing claimants. A rational analysis of the consequences of tuition systems requires separation of two questions: (1) Who should bear the costs of education? (2) If someone other than the student should pay for his education, in what form should the aid be given? Unless the distinction between these two issues is grasped, confusion is inevi_table. The case for zero tuition is not established by demonstrating that aid to students is desirable. Full tuition may still be desirable, with the desired aid taking the form of explicit grants-in-aid or scholarships from which the student pays .the tuition fee of his chosen school. The issue of the most desirable form of aid should be separated from still another closely related question: What is the desired method of financing and controlling colleges-as distinct from financing students? For example, aid to students in the form of zero tuition means also that the state finances the colleges’ activities directly by legislative appropriations with the students and their parents having less influence on finan- The Impact of Free Tuition •205 cing and controlling the activities of colleges. Where student aid is in the form of grants-in-aid or scholarships, students and parents paying full tuition to their chosen colleges have a greater role in determining which colleges shall be financed and rewarded for superior performances. Recognition of these differences in effect explains why some people have asserted the administrators and members of state universities and colleges, which are currently financed by direct legislative appropriation, have sought from self-interest, rather than educational interest, to maintain the impression that zero tuition is the only feasible or sensible means of aid to students-in order to repress student influence and control over the colleges while retaining the influence of politicians. Advocates of subsidization of college students (regardless of the method) assume that if each student bore the full cost there would be too little college education as well as a decrease of educational opportunity. What makes it desirable to have more education than if students pay full costs? Several arguments are advanced. Let us discuss these in ascending order of sophistication. ( 1) ” Although the costs of education are less than the gains to the students themselves, some are unable to finance their education now. A subsidy would provide educational opportunity to the poor.” (2) “Cultural education, though not profitable in market earnings, and hence not capable of being paid for out of enhanced earnings, is nevertheless desirable.” (3) ”Even if every student acquires as much education as is worthwhile to him, he would take too little, because the individual ignores the beneficial _social gains indirectly conferred on other members of society-giving what some people call ‘external social effects.’ Therefore, society at large should 206• Economic Forces at Work induce students to take more education than indicated by their private interests.” The argument that the poor cannot afford to pay for a profitable college education is deceptive. What is meant by a ”poor” person. Is he a college-caliber student? All collegecaliber students are rich. in both a monetary and nonmonetary sense. Their inherited superior mental talent-human capital-is great wealth. For example, the college-caliber student is worth on the average about $200,000, and on the average, approximately $20,000-$50,000 of that has been estimated as the enhanced value derived from college training, depending upon his major field and profession. Failure to perceive this inherent wealth of college-caliber students reflects ignorance of two economic facts. One is the enormous human wealth in our society. Every good educator recognizes that inanimate capital goods are not the only forms of wealth. The second fact is the difference between current earnings and wealth. For example, a man with a million dollars’ worth of growing trees, or untapped oil is a rich manthough he is not now marketing any of his wealth or services. So it is with the college-caliber student. Though his current market earnings are small, his wealth-the present wealth value of his future earnings-is larger than for the average person. This is true no matter what the current earnings or wealth of his parents. It is wealth, not current earnings nor parent’s wealth, that is the measure of a student’s richness. College-caliber students with low current earnings are not poor. Subsidized higher education, whether by zero tuition, scholarships, or zero-interest loans, grants the college student a second windfall-a subsidy to exploit his initial windfall inheritance of talent. This is equivalent to subsidizing drilling costs for owners of oil-bearing lands in Texas. The Impact of Free Tuition ·207 The_re remains an even more seriously deceptive ambiguity-that between the subsidization of college education and provision of educational opportunity. Educational opportunity is provided if any person who can benefit from attending college is enabled to do so despite smallness of current earnings. Nothing in the provision of full educational opportunity implies that students who are financed during college should not later repay out of their enhanced earnings those who financed that education. Not to ask for repayment is to grant students a gift of .wealth at the expense of those who do not attend college or who attend tuition colleges and pay for themselves. This is true because, for one reason, our tax bills do not distinguish between those directly benefited by having obtained a zero-tuition educational subsidy and those not so benefited. Alumni with higher incomes pay more taxes, but they do not pay more than people with equal incomes who financed their own education or never went to college. Many discussions about educational opportunity refer to proportions of students from poorer and richer families at tuition-free colleges. However strong the emotional appeal, the proportion of rich and poor family students is relevant only to the separate issue of wealth redistribution, per se, consequent to state-operated zero-tuition education. It has nothing to do with the extent of educational opportunity. Though data for California colleges and taxes suggest that lower-income groups provide a smaller proportion of students than of taxes to support education, such comparisons are irrelevant, so far as provision of educational opportunity is concerned. These data tell how much wealth redistribution there is among the less educated, the poor, the educated, and the rich. That wealth redistribution is good or bad depending upon whether f 208• The Impact of Free Tuition Economic Forces at Work one believes the educational system should be used as a device to redistribute wealth as well as to enhance wealth, knowledge, and educational opportunity. No matter how zero tuition in tax-supported schools may redistribute wealth, the provision of full educational opportunity does not require redistributions of wealth. Yet, it seems to me, many people confuse these two entirely separate issues or think the latter is necessary for the former. To think that college-caliber students should be given zero tuition is to think that smart people should be given wealth at the expense of the less smart. When some zero-tuition university alumni say that without zero tuition they could not have attended college, they should have a modest concern for the implications of that statement. One poor, “uneducated” resident of Watts, upo,ri hearing Ralph Bunche say that he could not have had a college education unless tuition were free, opined, “Perhaps it’s time he repay out of his higher income for that privilege granted him by taxes on us Negroes who never went to college.” That reply spots the difference between educational opportunity and a redistribution of wealth. be provided if collegeFull educational opportunity would ._ . caliber students could borrow against their future enhanced earnings. Students could repay out of their enhanced future earnings. Although, currently, loans are ~vailable from private lenders and also from publicly supported loans, a subsidy could provide a state guarantee of, repayment of educational loans exactly as housing loans .are guaranteeed for veterans. Students could select among optional repayment methods. Some could contract to repay in full with interest; others could opt for a sort of insurance system, whereby the amount repaid was related to their income, with upper and lower limits to amounts repaid being specified. A host of possibilities are I .·,• 1 I •209 available. In fact today with income taxes, the college alumni are repaying part of the educational costs via taxes (but so arc others who did not attend college). Some people are impressed by the size of the debt that a college graduate would have to repay, but they should be impressed with the fact that the debt is less than the enhanced earnings he has thereby obtained and is an indication of the wealth bonanza given the student who is subsidized by society. There remains one more facet of the educational opportunity argument. Even if a college education may be a very profitable investment for some person, he may, because of inexperience or lack of confidence, not appreciate his situation or be willing to borrow at available rates of interest. This presumably is an argument for subsidizing those students who lack confidence or understanding of their possibilities, and it may be a meaningful argument on its own ground, but it is not an argument for subsidizing “poor” students. Pleas are made for subsidizing cultural education which, though it may add nothing to the student’s future market earnings, will enhance his general welfare. But a person’s welfare is increased if he gets more food, housing, recreation, beer drinking, and fancier cars. It would seem therefore that the . relevant argument for helping students is one of helping them regardless of whether they wish their welfare increased via cultural education or better food. A grant of money to be spent as the recipient deems appropriate is an efficient form of aid-as judged by the recipient. Subsidized cultural education rather than money gifts could be justified if the giver knows better than the recipient what is good for the recipient. I cannot make that leap of faith for the collegiate student, although other people do it easily and confidently. 210• Economic Forces at Work A case can be made for subsidizing the poor and the rich to take more education-more than a person would take when motivated by his own interests alone. It is often said there are privately unheeded, net social benefits, so each person will underinvest in education from the social point of view, regardless of whether he is rich or poor; but we must separate the illusory from the real external available gains. Education makes a person more productive, as a doctor, lawyer, merchant, or engineer. Other people benefit from his greater productivity, because more engineers enable lower costs of engineering services for the rest of society. Engineers, looking only to their private gain would, it is said, undervalue the total benefit of having more ep.gineers; too few people would seek sufficient engineering’–‘education. If this sounds persuasive, economics can teach you something. The increased supply of engineers reduces the prices of engineering services-even if by only a trivial amount-and thereby reduces the income of other engineers. Their income loss is the gain to the rest of society. This is a transfer of income from existing engineers to nonengineers; it is not a net social gain. The benefited parties gain at the expense of existing members of the engineering profession, who lose some of their scarcity value as more educated people are created. This is a transfer from the more educated to the less educated. A striking awareness of this effect is evident in the advocacy by labor groups of immigration restriction. Restricting the inflow of laborers of particular skills prevents reductions in wages of incumbent workers with similar skills and prevents a transfer of wealth from them to the rest of American society. An immigrant or a more educated person would have provided an increased produr,t and he would have obtained that value by the sale of his The Impact of Free Tuition • 211 services, but the lower wages to that type of services would have transferred some of the incomes of similar workers to the rest of society. This external transfer effect is not a net con-· tribution to social output. It is not a reason for subsidizing education. For external effects to serve as a valid basis for more education two conditions must be satisfied: (1) There must be a net social gain (not transfer) unheeded by the student. The ability to read reduces dangers and inconvenience to other people; ability to be sanitary enhances health of other people, or economic education may-but probably will not-prevent . passage of socially detrimental, special-interest legislation. These are examples of education with external social gains, which we shall assume are not heeded J:,y the student in his private actions because they do not affect the marketable value of his services. Professional education of doctors, engineers, lawyers, economists, mathematicians, etc. , has not been shown to fit in that category. Perhaps education at the undergraduate collegiate level in the elements of law, psychology, political science, mathematics, economics may make for better nonmarket decisions or actions. I confess to a strong suspicion that such education is most significant at the grade school level, diminishes at higher levels, and disappears for professional or cultural, artistic, personal satisfaction courses, and is possibly reversed at graduate levels (by overtraining and insistence on excessively high standards of training for granting of licenses to practice in some professions-though this is a point the validity of which is not crucial to the main issue here). (2) The second condition is that there must be further external gains unheeded by students at the college level. The fact of having achieved net external gains is not sufficient to warrant 212 • Economic Forces at Work The Impact of Free Tuition • 213 subsidization. The crucial condition is the failure to achieve still further available incremental net social gain fromfurther education. Before concluding that they exist because of a tendency for people to ignore them, we should note that people· attend college for reasons other than financial marketable gain. College attendance for personal reasons includes cultural, artistic education, and attendance to find mates. All these tend to extend education beyond maximizing one’s market wealth and possibily even beyond that yielding unheeded social gains. But the facts are not conclusive in either direction. Incidentally, an especially common but erroneous contention, presumably relying on the external effect, is that the growth, prosperity, and unusual position of California depend upon the free-tuition, higher education system. What does this mean? If this means that free tuition has contributed to higher wealth for the educated then this is no argument for either free tuition or more education. If it means the prosperity and growth of aircraft, electronics, motion picture, or agricultural industries in California are dependent upon f.ree tuition, the contention remains unsupported by any analytic or factual evidence, and in fact can be falsified by_ comparisons with other states. Even if it could be demonstrated that subsidized higher education was responsible, the issue of free tuition would still not be touched. If this means that free tuition did attract some people to_ seek their education in California, they proceeded to reap the gain in their own higher income. If they provided a real net social benefit, it should have exceeded the extent of their subsidization to be justifiable. The same proposition holds for residents of California. If this argument is accepted, it is difficult to justify charging newcomers a full tuition while permitting existing residents a “free tuition.” Yet, we 11.ave seen no proponent of zero tuition advocate zero tuition for all newcomers from all other states. If this means that the higher incomes for more people increase tax receipts, then the relevance of that completely escapes me. If this means California has a larger population, then this means higher land prices. But in so far as benefits to “California” have any relevance, I believe they should be viewed as benefits to people in California rather than as benefits to owners of a geographically identified piece of land, unless by ”California” one means “landowners or politicians,” who indeed do prefer larger populations as a source of political power and higher land values. To induce students to take more education than is privately worth their while-in order to obtain the otherwise unheeded external gains-does call for payments to students. If a student were paid for doing what he would have done anyway, or if h!s education were subsidized to increase his wealth, he would l;?e receiving a gift. ~_pay_~~-1:LCw-hether _:;t.§ Z~fo tuiti2n_.0r_a_ ~1?:~ypayment) to the student to extend his education, for the sake of achieving realJ _external be@fits that !?-~the~j,§,~ would__Q.av~not_pLQ.duc.ed,is_)t_Q~)’ment for services, muc~__as ~—~–···–····—~–~~……. .if build houses, for the benefit of the rest of society. Such payments may well be independent of the income or future income of the student as well as of his parents. Though there is nothing that says the rich would provide less real external effects from more education, my conjecture is that the rich would in any event take more education than the poor for cultural reasons and would therefore require a smaller inducement to take the “optimal” extra amount of education for external social benefits. This can form a basis for advocating more educational inducements to the poor than to the rich, but not necessarily by a zero-tuition inducement to rich and poor alike. – 214· Economic Forces at Work It should be noted however that there is already subsidization of higher education by private philanthropy on a scale that staggers the imagination. The endowment funds of colleges and philanthropic foundations aiding education run into the· scores of billions. Even if only half that were used to subsidize education (and the rest for research), the amount can not be regarded as minor, on any standard. No matter what your beliefs about the validity or relevance of the preceding consideration, let us accept them, for the sake of analysis of alternative means of providing aid, for full educational opportunity, cultural aid, or extra inducements to education. (Of course, those who think the preceding arguments are too weak to warrant taxpayers’ giving aid to college students can ignore all that follows, for to them there is no case for any state action, nor of zero tuition.) The rest will want to ask, “What is the best form of aid or inducement?” We can enable or induce students to take more education with the following offer: “On the condition that you take certain kinds of education, we shall bear enough of the costs to induce you to do so.” The costs he would hav.e borne are the income forsaken and the tuition costs. (Food and living·costs can be ignored for he would be incurring them no matter what he did.) Which of the following is the preferred way of extending that aid to potential students? (l)’We pay directly the costs of extra education by operating the school to provide the extra education; this is the zero-tuition system. (2) We pay him an equal amount on the condition he take the additional, specified type of education, but he decides which school to attend and he pays the tuition to the school. This is an educational voucher or G.1.-type educational bill-of-rights (used after World War II for veterans). The Impact of Free Tuition • 215 The first requires also that the state directly finance and operate the school providing the education; the second permits the student to choose from competing schools and direct payment to the school he chooses. These two alternatives are sufficient to illustrate the major implications of zero versus high tuition modes of subsidy. The wealth effect for the student is superficially the same in either case, and the financial cost to the subscriber can be the same in each case, once it is decided how much education to subsidize for whom. The costs to the subscriber may be the same, but the results are not. In the California state system of higher education, the tuition fee is zero.for all state schools and for all kinds of training, regardless of whether it contributes to a net social gain or not, and regardless of how rich the student is. Zero tuition implies that the appropriate aid or subsidy for every student of a state school is exactly equal to the tuition cost no matter what subject he takes. No basis for zero tuitions as being the proper .amount has ever been presented; maybe the aid should be even larger, to compensate for forsaken . ‘ earnmgs. Because low- or zero-tuition schools are believed to have a larger proportion of less wealthy students than high-tuition colleges, zero-tuition schools are believed to do a better job of providing educational opportunity for less wealthy students. But this entails the earlier confusion between provision of opportunity and provision of a wealth bonanza; zero-tuition schools give bigger wealth gifts to the mentally able students than do the high-tuition schools. Of course, higher tuition will, other things left unchanged, reduce the number of financially insecure students attending tuition colleges. The case for raising tuition is not that aid should be denied but instead that “zero-tuition” is a less de- 216• Economic Forces at Work sirable means of providing aid to students; it entails undesirable controls and political interference with education and lowers the quality of education. Yet there is another method of providing full educational opportunity and at the same time improving the quality and quantity of education and reducing political controls. The alternative is a system of full tuition supplemented by grants-in-aid to those who qualify as financially insecure and deserving students. It is important to note that the financing of colleges to provide education is different from subsidizing students. The zero tuition is a subsidy to the college as well as to the student. Subsidies to students alone can be provided with a full-tuition system: in fact they are now being so provided by many private schools that do charge full tuition. The alternative to the zero-tuition method of providing educational opportunity or giving aid is tuition, with loans or with grants of money. The critical difference, in my opinion, between no tuition and tuition, under these circumstances, is that the former lets the state politician and college administrator and faculty directly exert more control over education whereas the latter enables the student to exercise more power by his choice of college. Subsidies to whatever extent desired could be provided by a· system of grants-in-aid via scholarships. That would appear to be more expensive administratively (but on ly administratively) than zero tuition, precisely because an effort is made to eliminate the haphazard bonanzas in the zero-tuition system. The presumption is that the cost of selecting the students to be subsidized is less than the savings from the avoidance of subsidies to all students. Tuition with grants-in-aid to students is not visionary. It is proven, practical, economical and currently used. New York 1 The Impact of Free Tuition ·~· I)’, ·217 State already has a large system of Regents scholarships. California has a smaller scale system with about 2,000 scholarships. After World War II, the federal government granted millions of veterans educational vouchers for tuition, books, and incidental expenses under an enormously successful act known as the G.I. Bill. All these granted aid regardless of the student’s current financial status. In California the university and state colleges now receive about $500 million annually directly from the legislature. That would finance 250,000 scholarships of $2,000 each. The university’s budget would finance 125,000 students, more than the number now attendmg. At present many arrangements exist whereby private colleges take into account the financial status of students in deciding how much tuition to charge each student. Even more efficient would be a system ofloans with interest to be repaid after graduation out of the student’s enhanced earnings. Under a loan system, the problem of filtering rich stu?ents from the financially distressed would be reduced to trivial dimensions, since the rich would have little, if anything, to gain by borrowing. This would provide full educational opportunity with little need for a means test. Full tuition does not in any way restrict the achievability of full education opportunity. That can be achieved explicitly and openly by the scope of grants and subsidized loans. Just as social security and welfare payments are made in money with the recipient choosing his purchases from competing producers, so a full-tuition system with grants-in-aid or loans would enable separation of the issue of the amount, if any, of the subsidy from that of the best means of providing and controlling education. Under a system of full-tuition fees, with whatever loans and 218• Economic Forces at Work scholarship voucher grants are deemed desirable, students could choose their education from the whole world. Any accredited college or educational institution whether it be for barbers, television technicians, beauty operators, mechanics, butchers, doctors, lawyers, or historians could serve. Ours would then really be the best educational system in the world; no longer would Californians be confined to California stateoperated schools. Whatever one’s beliefs about the desirable degree of subsidy for more education, and whatever his beliefs about who should get it, the full tuition voucher coupled with scholarships and loans would magically open a new, larger world of choice. An alternative form of aid to students is a tax-credit allowance whereby parents, or students, could later receive a tax offset to their payments for tuition. This would put private college students on a more equal basis with low tuition public colleges. In my opinion, this would be equality at the wrong level of equality. Rather than give tax credits as a means of maintaining zero tuition, I would prefer placing a tax liability on students attending public colleges with low or zero tuition. Whereas the tax credit provides subsidies and aid to all students at the expense of nonstudents, the tax-liability assessment places the costs of providing the education more squarely on those who benefit from the education. A t~x credit gives equal treatment to private and public college students-at the expense of nonstudents. A tax liability gives equality to private and public college students and to college and noncollege people, with each bearing only the costs of service provided for their benefit. If tax-liability assessments are out of the question politically, the tax credit would be the next best; but it would not achieve one of the major purposes of a full tuition system. The Impact of Free Tuition • 219 With full-cost tuition, competition among California colleges, and even among academic departments would change. Instead of competition for funds being negotiated among university committees, deans, regents, state college boards, and legislators, competition would rely more on classroom behavior of instructors who would be more dependent on student attendance vis-a-vis other departments and other colleges. This would enormously enhance the power of the student in the former zero-tuition colleges. Giving students more attention and influence in the university would indeed occur, exactly as the customer exercises more power at the grocery-by his purchases and choice among competing products and stores, but not by leaping over the counter and insisting on power to run the store, as occurs with current protest. Currently at the grade school level many parents are turning to private schools precisely because the parents can choose more fully the kind of education given their children-via the power of the purse. The poorer people do not have that option-but they would with a tuition-grant system. Since the producer usually knows more about what he is producing than does the consumer, the producer illogically tends to conclude that he is a better judge about the appropriate quality and quantity for the consumer. This tendency is especially rewarding if the producer can thereby obtain a sheltered competitive position in the production of the good. He would tend to produce a quality and quantity in a style related more to that which enhances his welfare and less to what students and parents prefer. It is easy to see that with zero tuition the university faculty benefits from research and graduate activity that builds an impressive publication record and research status, with the currently less rewarding teaching of undergraduates being re- 220· Economic Forces at Work legated to the less “distinguished,” lower-ranking faculty or graduate students. The “publish or perish” rule would be less powerful under full tuition, because teaching would become a more important source· of student directed funds. Survival of the better teachers who are weak in publication would be enhanced. It is interesting and amusing to note, incidentally, that students at the University of California are now attempting to protect some members of the faculty from being dropped because of inadequate research and publication. The protection comes by the students ”donating” funds to hire the man to give classes; this is a voluntary, spontaneous full-tuition system. If allowed to expand, students would determine who was on the staff and who got the bigger incomes, just as they now decide which restaurants shall survive and prosper. This is a simple application of the old, powerful, fundamental principle of behavior. The lower the price at which goods are distributed, relative to the market value, the greater the degree of discrimination and arbitrary criteria that the ”seller” will display. Its corollary is that the lower the seller’s right to the monetary proceeds, the greater his gain from underpricing the goods. The gains to the university administration and faculty from low tuition are classic examples, first expounded in Adam Smith’s The Wealth of Nations. The greater the portion· of a college’s funds coming from tuition fees, the ‘greater the power of the students and the greater the role teaching will play in the survival and prosperity of the members of the faculty. The less will the faculty choose which students shall’ attend, how they shall behave, etc. The lower is the ratio of tuition payments, the greater the power of the faculty over the students because the students are less able to exert significant effects on the financing of schools or departments as a reward for “good” performance-as they can with restaurants. The faculty says The Impact of Free Tuition • 221 ”education is different” and students are poor judges of good education; students are swayed by popular, theatrical teachers and do not appreciate the more valuable scholarly teachers. One wonders how students happen to go to the better and possibly tougher schools in the first place. The faculty of any college prefers lower tuition-until the budget expenditures can not be met from nontuition sources. And even then there is conflict of interest within the college between those who are threatened by the budget cut and those with tenure who are not. If the cut, or loss of income, would mean merely fewer undergraduates and fewer new teachers, clearly the least difficult resolut~on from the current faculty’s interest is the reduction in new students, rather than an increase in tuition. With zero tuition the state schools have expanded relative to higher-tuition private colleges, and the state university with its higher-salaried teachers and more expensive education is more attractive to students than the state colleges and junior colleges. The ex-president and the administrators of zero-tuition institutions correctly insist that zero tuition is the great principle underlying the growth of the university; but it is not a source of better education for California students. We should not confuse the amount of money with the way the money is obtained. More and better education, as judged by students, could be obtained at the same, or less, cost with the full tuition control of colleges coupled to loans and whatever grants-in-aid are desirable. With full-cost tuition, the less expensive junior colleges would attract students and income from the university and colleges. Predictably, the few administrative voices heard in favor of higher tuition seem, from my observation, to come from junior college administrators-who believe they would 222 • Economic Forces at Work outperform the university if put on a quality-cost basis of competition for students. A counter argument to the preceding propositions is that junior college education is ”inferior” to university education. Although the quality of the university as a research institution is high, not as much can be established for its quality as a teaching institution to educate college students. The move to junior colleges with full tuition would occur if the more expensive university education were not matched by the higher quality as judged by students and parents. The university would have to improve its teaching to hold students at its higher costs. If it could not, the results would constitute evidence that the high-cost and high-quality combination was not a superior combination of quality, cost, and quantity. A Rolls-Royce gives higher-quality transportation than a Ford, but it does not follow that more Rolls should be produced than Fords. Education must be judged by the quality, quantity, and costs, rather than in terms of only those who are educated at the highest, most expensive levels. Yet, despite this patent fact of life, when faced with a budget cut the administrators of the state university plump four square for ” quality at all costs’ ‘-for maintenance of quality education for a selected few regardless of how many must be· turned away and given instead an “inferior” education. On what criterion is it established that it is better to maintairi the level of quality of education for fewer .students at the cost of . sacrificing education for others? Would one argue that in the event of a social security reduction, we should reduce the number of recipients in order to maintain the quality of those lucky enough to keep getting social security payments? But analogies aside, the elite, authoritarian ~rguments by university administrators and faculty for a given level of quality, The Impact of Free Tuition • 223 regardless of the sacrifices imposed on excluded students or on taxpayers, are sobering evidence of the seductiveness of selfinterest pleading. imJ The faculty and administration of higher education in California have evolved in the zero-tuition environment, with appropriately adapted behavioral traits. They have learned to use that political structure; they have learned how to appeal to the political processes and to legislators and governors for more financing. They have been almost exclusively reliant on the political process. They praise politicians for statesmanlike, . responsible behavior when the university budget is increased; but if it is decreased, they cry of political interference. Having accepted almost exclusive dependence on financing directly from the political and legislative processes, they should not complain of “political interference” when that same political process examines more intently the budget and the operations of the university. Are they really surprised that the venerable law “He who pays, controls” still is effective? Legislators generally tend to favor direct state legislative financing of education coupled with no tuition, rather than full tuition with grants-in-aid. The closer the tuition approaches full cost, the less the power of the legislators over the educational institutions. It is not entirely accidental that Congress used a grant-in-aid system for veterans; there was no federal college system. We must constantly remember the difference between paternalism and independence. Independence from the competition of political processes and politicians’ interests can be enhanced by full tuition, but it will bring greater dependence on competition among educators in satisfying students’ whims and interest. Either the students pay and control, or the politi- 224• Economic Forces at Work· The Impact of Free Tuition ·225 cal processes and politicians do. Yet some of the faculty seem to think they can avoid both. For educators there is no free lunch nor “free” tuition. The situation reminds one of the Russian plight. Dissatisfaction with the quality of goods produced by Russian firms is sparking attempts to restore market prices as reflections of consumers’ interests. While the Russian economists and consumers advocate more control via the market, producers and politicians show far less interest in weakening their power by moving away from socialism. There remains a subtle, but effective means whereby full tuition would lead to more education than if directly provided by government at zero tuition. As matters stand now, an education at a tuition school may be worth $2,000, or say, $500 more than the education at zero-tuition state schools. For that superior education worth $500 more, the student would have to pay the full-tuition cost of $2,000. He gets no relief for not using state schools. If education were on a full-tuition basis, this obstacle to more and higher quality education would be removed. We should not assume that spending more by government for direct provision of education necessarily yields more education. This phenomenon, I conjecture, is powerful at all levels of education. A preference for full tuition implies. nothing whatsoever about the desirable extent of aid or subsidy to students. Unfor-. tunately much of the debate has erroneously assumed that zero tuition is a necessary or a preferred method of aid while full tuition is a device to avoid aid to students. No matter how much aid, if any, should be given to students, the case for full tuition does not rest on a denial of aid. 1t rests on the premise that, whether or not aid is given to students, the financing of schools should be controlled more directly by students and their parents because the kind of education thereby made available is deemed to be better-by those who advocate full tuition. Full tuition, plus grants-in-aid to whatever extent one believes is justified, directs educational activities more to the interest of students and less to that of the university staff. And after all, is it not the students whose interests are fundamental rather than the university’s as an institution? Is it the students’ interests as reckoned by students and parents rather than the convenience to the educators that is a better guide? My choice of answers is obvious. I suspect that these are the crucial issues on which advocates of zero tuition will differ with me. My opposition to zero tuition arises because I do not like the way it redistributes wealth, nor do I like the totality of the effects of the kinds of competition it induces relative to that which would prevail under full tuition, supplemented by grants and loans. The latter yields more variety of educational opportunities and just as much educational opportunity and presumptively, greater detectability and survival of superior education. It reduces the producers’ control over the products that the customers can have. The influence of selecting their colleges and controlling payments is a trait with high survival in the world outside of academia and which should be cultivated. The decreased role of the state and political activity in administering education is also a consequence I find congenial. Higher tuition would improve the quality of education rather than reduce it. The quantity would be affected not by either a zero or a high tuition, but by how much is spent for education. Zero tuition does not mean more is spent for education, nor that more poor people can attend. To believe it does is to think zero tuition is the only or best way to subsidize or 226· Economic Forces at Work , ”\’ aid students-and that contention begs the fundamental question of what is the best way. All these consequences seem to work against my interests as a member of a zero-tuition college. If I thought this one exposition of economic analysis and one man’s preferences really were capable of converting our system of educational subsidies from the zero-tuition to a full-tuition system with scholarships, loans, and vouchers, I might be less willing to expose it, for the price may be high enough to make me join with those who, whatever may be their reason, prefer the Holy Zero (excuse me, the free) tuition system. Who Pays for Free College? Crowding Out on Campus∗ Alonso Bucarey† Job Market Paper – Latest version at: http://economics.mit.edu/grad/bucarey/research January 16, 2018 Abstract Free college tuition has been central in the higher education policy debate. In Chile, a government elected in 2014 promised free university tuition by 2020. Most research on tuition subsidies focuses on enrollment gains for newly eligible students. This paper studies spillover of these policies to students currently receiving generous financial aid. I show that free tuition increases demand at selective programs, making these programs more competitive and pushing them out of reach for many low-income students who would have qualified otherwise. The argument uses a combination of reduced-form regression-discontinuity estimates of enrollment elasticities and a structural model that captures general equilibrium effects. Estimates using Chilean administrative records suggest that 20% of currently enrolled poor students will lose seats to wealthier students under a free-tuition policy. This adverse effect on low-income students could be mitigated by complementary policies such as capacity investments and means-testing. However, crowd-out remains significant unless aggressive policies to counteract it are enacted. ∗ I thank Nikhil Agarwal, Joshua Angrist, and Parag Pathak for their invaluable guidance and support. I also thank participants of the MIT Labor field lunch for their useful suggestions. Special thanks go to Marcelo López and the Ministry of Education of Chile for assistance and data. This work was supported by a National Academy of Education/Spencer Dissertation Fellowship. † MIT Department of Economics. Email: bucarey@mit.edu 1 1 Introduction Making college tuition free for all has been at the center of policy debates in several countries and states. For example, Chilean students have protested since 2011, demanding that higher education become more affordable. In 2014, Chileans elected a new government that promised to make college free for everyone by 2020 (Bachelet, 2012). This is also a contentious issue in other countries, and in each of these contexts, advocates argue free college would expand access to higher education for groups left behind by financial aid policies (Clinton, 2015; Sanders, 2015).1 This debate has largely ignored the potential negative consequences that free college tuition might pose for low-income students who already have access to higher education as a result of financial aid. An across-the-board reduction in college costs may cause middle- and upper-income students to apply to colleges that are more expensive. At selective institutions, these new applicants might end up crowding out low-income students, who on average have lower test scores. The equilibrium effects of free college on admissions and its distributive impact depend mainly on three factors: the change in preferences over college alternatives, the distribution of admission test scores and income among students, and the socio-economic composition of students who are at the margin of admission. Because prices do not clear the market, programs that face an excess demand will therefore need to become more selective in their admissions. In equilibrium, students who are crowded out are either directly displaced by new beneficiaries of financial aid who changed their applications or displaced by the previous group after they try to enroll elsewhere, starting a chain of displacement. It is challenging to study the phenomenon of crowding out because free college affects the whole market. In this paper, I use rich administrative data from Chile to study the access to higher education for low-income students under free tuition and the design of financial aid policies. College students in Chile are directly admitted into a specific college-major, and admission depends exclusively on a weighted average score of a national admission test and students’ high school GPA. Each college-major admits the best applicants according to a score, generating a transparent channel for crowd-out: a more qualified student can displace students in the admission process. Additionally, financial aid is awarded based on sharp eligibility rules on income and test scores. Finally, financial aid expansions that took place since 2012 1 Countries where this has been a topic of discussion in recent years include Chile, England, Germany, and the United States. In the US, the states of Florida, Maryland, New Jersey, Ohio, Washington, and Wisconsin have introduced proposals to make tuition free for all, while other states and cities have already implemented free tuition policies, including Chicago, Illinois; Detroit, Michigan; New York; Oregon; Rhode Island; San Francisco, California; and Tennessee. 2 provide the opportunity to study whether low-income students were crowded out. The argument is developed in three parts. First, I test whether low-income students were crowded out by using past expansions in financial aid eligibility. Until 2011, eligibility for government scholarships was restricted to students in the first two income quintiles, and starting in 2012, it expanded to cover students in the third income quintile. I use this change in a difference in differences strategy and compare college-majors with different shares of students with a scholarship in the pre-expansion period as a proxy to where new beneficiaries would apply. I find that programs with a one standard deviation higher share of low-income students in the pre-expansion period experienced an increase in admission cutoffs of 0.7 standard deviations and a 10% drop in their share of low-income students after financial aid expanded. However, the causal interpretation of these results is complicated by the equilibrium nature of the problem. Students who are crowded out might displace others when applying to their next-favorite alternative. This propagation has the potential of affecting all programs and control groups. For this reason, and in order to study the distributive implications of free college and the importance of its design, my main empirical strategy directly addresses these forces. Second, I build and estimate an equilibrium model of the college market. The model allows me to determine how demand changes with free tuition, how selectivity would need to increase in order to clear the excess demand, and who is crowded out in equilibrium. To build and estimate the model, I leverage the existence of a large centralized admission system and the discontinuous eligibility for scholarships. The centralized admission system in Chile uses a version of the deferred-acceptance algorithm that creates a stable match between students and colleges.2 Stability implies that everyone chooses their favorite program within their choice set, and I use this to estimate a standard random utility model that allows me to determine each student’s ordering of college-majors under different financial aid policies. Additionally, knowing the rules used in the admissions process simplifies finding the new match under free tuition. The key parameter of the demand model, the sensitivity to prices, is estimated using a regression discontinuity (RD) design on scholarship eligibility. The RD provides local but large quasi-experimental variation in tuition that helps to identify the key parameter for the free tuition counterfactual and solves the price endogeneity problem of demand estimation. Building on Berry and Haile (2014), I prove that the RD non-parametrically identifies price response for a general class of preference models. Additionally, reduced form RD analysis reveals that scholarship eligibility increases enrollment by 6 percentage points and motivates students to substitute away from non-eligible programs. This is in line with a growing quasi2 The centralized admission system enrolled 76% of new high school graduates in 2015. 3 experimental literature on the effects of financial aid for newly eligible students in different countries and regions (e.g. Goodman (2008), Cohodes and Goodman (2014), Angrist et al. (2014), Goldrick-Rab et al. (2016), and Solis (2017)). Third, my equilibrium analysis shows that when college is free but capacity is unchanged, low-income students are crowded out. They are hurt because middle-income students would apply to more selective and expensive colleges, displacing low-income students. Indeed, my demand estimates show that under baseline admission cutoffs, free tuition would create an excess demand of 36%. In absence of prices, this market adjusts any excess demand by increasing selectivity; this increase disproportionately affects low-income students, who are more likely to be the least qualified admitted students. I estimate that around 13% of students in the poorest 40% of the population would be displaced from programs at the centralized assignment system, and around 5% of them would end up unassigned after free tuition is introduced. Consequently, the policy has a negative welfare effect on low-income students, with losses ranging between 3 and 6 thousand dollars, while producing small gains between 0 and 1.5 thousand dollars for high-income students. The scenario where college is free and holding capacity is fixed is a relevant short-run benchmark because capacity expansions might be costly, and this policy mirrors the proposed government plans in Chile and elsewhere. Finally, in practice, efforts to expand financial aid might be accompanied by colleges capacity expansions or more elaborate means-testing schemes. I therefore analyze these quantity and price instruments to determine what is needed to avoid crowding out lowincome students. I show that college capacities need to increase by 10% to maintain the enrollment rate of low-income students at baseline levels. For colleges participating in the centralized system, this expansion would be twice as large as the growth experienced in the last decade. Moreover, colleges need to expand their seats by more than 20% to ensure free college does not hurt a single low-income student. This is similar to the total expansion in enrollment in the last decade among all universities, but it is unseen among the more traditional and prestigious programs at the centralized system. Alternatively, to preserve low-income students’ enrollment rate while holding capacities fixed, the government would need to means-test scholarships, with the poorest 20% receiving a full scholarship and a decrease in the amount of the benefit with income, up to a scholarship of 50% for the wealthiest 20%. Chile is an ideal setting to study my questions for several reasons. First, there is a centralized assignment mechanism that matches students to their favorite college-major using a mechanism based on Gale and Shapley’s (1962) student proposing deferred acceptance algorithm, which produces a stable outcome. Further, selective universities outside the 4 centralized system base their admissions only on GPA and test scores. Second, there are rich administrative micro data for the whole higher education system, financial aid and, national test scores. Finally, like other countries, the issue of free college and expanding financial aid has been at the top of the policy agenda for many years. My study relates to several strands of the literature. Bound and Turner (2007) document crowding out in the context of larger cohorts in the US. Their focus is on the lack of capacity expansions among colleges caused by reduced non-tuition resources per student. Murphy, Scott-Clayton, and Wyness (2017a, 2017b) provide descriptive evidence that low-income students may have been hurt by increased competition during England’s free tuition era. These articles do not decompose resorting of students and the distributive consequences of crowding out, which is my focus. My empirical analysis aligns with research on the role of prices and quantities in centralized matching markets (Agarwal (2015, 2017)) rather than demand itself or the assignment mechanism (e.g. Hastings et al. (2009), He (2012), Agarwal and Somaini (2014), Fack et al. (2017), Abdulkadiroglu et al. (2017), Kapor et al. (2017)). Studies of financial aid such as those by Cohodes and Goodman (2014) or Angrist et al. (2017) primarily focus on identifying the price effects of college,3 while my study uses this idea as a foundation for an equilibrium analysis. Other work such as Kapor (2016), Fu (2014), and Arcidiacono (2005) construct equilibrium models of higher education, but their analysis is not typically built from quasi-experimental variation. The rest of this paper is organized as follows. The next section describes the institutional setting and data used in this project. Section 3 shows quasi-experimental estimates on students’ price responsiveness. Then, section 4 presents a stylized model to interpret the role of financial aid on enrollment decisions and evidence of crowding out from past financial aid expansions. Then, section 5 presents the empirical strategy, the model of equilibrium, students’ preferences, how the RD identifies price elasticity, and estimation details. Section 6 presents the parameter estimates, and section 7 provides the counterfactual results. 2 Institutional Setting and Data This section describes the Chilean higher education system. Two institutional features are key to my empirical strategy: i) the most selective programs admit students using a centralized system and admission at other programs relies on GPA and test scores, and ii) scholarships are awarded using a discontinuous rule of assignment. This section also describes the data set used in this paper. 3 See Deming and Dynarski (2009) for a survey. 5 2.1 Higher Education Programs The Chilean higher education system has similar enrollment rates and a similar mix of public and privately owned institutions to the US.4 Students can enroll in three types of institutions: Universities, Professional Institutes, and Technical Formation Centers. Universities focus on 4- to 5-year degrees, and the other two types of institutions focus on 2- to 3-year vocational degrees. Total first-year enrollment in higher education stabilized in recent years (CNED, 2017). As a result, market dynamics in enrollment have a minor role in 2015, my current year of study. Between 2005 and 2011, first-year enrollment increased from 200 thousand students to 340 thousand (CNED, 2017). Bordon, Fu, Gazmuri and Houde (2016) study this expansion in detail, showing that between 2010 and 2015 six elite universities expanded their capacities on average by five seats per program, while the rest of the universities experienced a net reduction of one to four seats per program over a baseline of 46 seats per program. 2.2 College Admission Systems in Chile Admissions are based on high-school GPA and a national admissions exam. High school graduates can register once per year to take the national admission test (Prueba Selección Universitaria, PSU), an SAT-type exam with four sub-tests: Mathematics, Language, and the choice of taking Science or History. After receiving their scores at the end of the year, students choose institution-major combinations and can apply for admissions in the two separate systems described below. Table 1 summarizes the number, total enrollment, selectivity, and average tuition of programs in 2015 by type of institution and admission system. The centralized admission system hosts programs that are between one and three thousand dollars more expensive than the average program, and the least qualified admitted student in this system has one standard deviation higher score compared to her counterpart at the decentralized universities. Additionally, 25% of students enrolling in the decentralized system do not take the national admission test. The centralized admissions system groups 33 out of all 58 universities in the country.5 It accounts for 76% of first-year university enrollment among same-year high school graduates.6 4 See Solis (2017) and Hastings, Neilson and Zimmerman (2016, 2017) for direct comparisons between both systems. 5 Some form of coordination using an exam in the admission process has existed since 1967. 6 47% of first year enrollment at all institutions. Corresponds to 26% among all first year students in 2015, and 57% among all first-year university students in that year. The difference between these percentages corresponds to older students that enroll years after graduating from high school. 6 It is also the most selective part of the higher education system in Chile, with programs requiring an average score in Mathematics and Language above the 45th percentile just to be considered in the admission process. The centralized admission system opens once every year after test results are announced, allowing students to rank up to ten college-major combinations. Programs rank applicants based on a weighted score that is a function of Mathematics, Language, GPA, and either history or science scores. Given students’ rankings, program-specific scores, and capacities, the system offers students a seat at most at one program in such a way that no student-program pair prefer each other and are not assigned together. This is done using an algorithm built on Gale and Shapley (1962)’s student-proposing deferred acceptance algorithm described in detail in Appendix A.2.7 This process creates a cutoff for admission at each program, which corresponds to the score of the least qualified admitted student. Rejected students are entered into the waitlist for that program. After students decide whether to take the admission offer, in a second instance of the centralized admission system, waitlisted students might be offered admission. The use of the student proposing deferred-acceptance algorithm guarantees that in absence of cost and a limit to the number of programs students can rank, and that weighted scores do not include ties, students are assigned to their most preferred option among programs where they would be admitted (Gale and Shapley, 1962).8 Three aspects of the centralized system might affect this. First, the system allows students to rank up to ten options. If binding, students would need to strategically choose what to rank, leading to the possibility that students are not assigned to their most preferred option among those where they are eligible. However, among all students, just 1.5% rank ten options and only 0.02% are admitted in their tenth option. Second, some institutions impose a limit to the last position where a student can rank them to be considered for admission. Two of the 33 institutions will only admit students that rank them fourth or higher. However, the median institution imposes no limit, and the average requirement is to rank the institution 9th or less.9 Additionally, 88% of students were admitted in one of their top three choices, so these restrictions are not binding to most students. Finally, the weighted score used by programs might include ties in the last admitted student, which the system solves by adjusting capacities to include all students with the same scores. However, this does not impose violations on stability (Rı́os et al., 2014). All other institutions admit students with no coordination and by directly receiving 7 Institutions also have special admission systems for athletes and international students, but they represent a small part of the admission process. 8 DA is also strategy-proof (Dubins and Freedman 1981, Roth 1982). 9 Appendix A.5 shows the list of requirements by institution in the centralized system. 7 requests for enrollment. Programs at vocational institutions are not selective, while programs at the 25 universities outside the centralized system usually impose an ex-ante cutoff for admission based on admission test scores (PSU) and high school GPA. Appendix A.3 shows examples of these requirements for two of these institutions. 2.3 Financial Aid In Chile, students can get financial aid from the government in the form of scholarships and a state guaranteed loan. Eligibility for these benefits is determined using cutoff rules for family income quintile and average Mathematics-Language admission scores. As a result, students are eligible only if they are below a certain income level and above a given score. Additionally, tuition subsidies are institution-major specific, with the average scholarship covering 80% of sticker tuition. Table 2 summarizes the scholarship programs for 4- to 5year programs with their amounts and cutoffs for eligibility. Higher education institutions also provide some financial aid, but there is no data on their award size or its importance for the system. However, considering that the main objective they have is attract high-achieving students, and that these scholarships usually last only for the first year of a four- to fiveyear degree, they are likely to play a minor role compared to the government financial aid programs. Students apply to all forms of financial aid by filling out a brief online form that selfreports their family income. This information is cross-checked with the Chilean tax office and is used to assign students into income quintiles.10 In this paper I focus on scholarships and how their targeting affects different groups of students. Figure 1 shows how scholarship eligibility expanded between 2011 and 2015. Over this period, students who were just below the eligibility cutoff for these scholarships were also eligible for the state guaranteed loan —that by 2015 could be used by students of all income levels whose average Mathematics and Language score was approximately above the 3th percentile. This means that students who by a small margin are not eligible for the scholarship could enroll in college using a state guaranteed loan, which Solis (2017) suggests eased credit constraints. 2.4 Data The Ministry of Education of Chile provided access to several data sets on students: enrollment, admission test scores, family income quintiles, scholarship assignment, and other 10 Students’ per-capita family income is compared to income thresholds, constructed from a national socioeconomic survey (CASEN), to determine income quintiles. 8 demographic characteristics. Each student has a unique identification number across all data sets. Enrollment information includes the exact program where the student enrolls, as well as characteristics of the program: type of institution, location, duration, posted tuition, accreditation, and area of knowledge. Student demographic characteristics come from a survey, administered to students during their registration for the admissions test, that includes parental education and type of high school students attended (public, voucher, or private), among other characteristics. My main analysis sample focuses on the cohort of students that graduated from high school in 2014 and immediately took the admission test (PSU) to enroll in 2015. This restriction ensures that the sharp discontinuity in scholarship eligibility is not subject to manipulation through multiple test taking behavior and avoids having to model the dynamic aspect of older graduates who might start college years after finishing high school.11 This sample contains 171,011 students, whom I observe in 107,693 bins created from a subset of student demographic characteristics (type of high school, mother head of household, mother’s education level, and region of residence), program of enrollment, income quintile, and financial aid status. For each bin, I also observe the average Math-Language score and average GPA. In this sample I closely replicate the regression discontinuity using the full data set, which makes me confident that the aggregation process is not a problem. The main restrictions of my current data are not having detailed information about individuals’ test scores and observing income quintiles instead of deciles. However, observing GPA and average Math-Language score closely approximates the actual admission choice set that students face, and the fourth income quintile has a minor participation in scholarships through the seventh decile. Further details on this issue are presented in Appendix A. Table 3 shows details for the population of students, their enrollment rates, and other observable characteristics used in estimation. In the data, I do not observe per capita income, but only the final income quintile used by the Ministry of Education to assign financial aid. There are 27% of students for whom I do not observe income quintile because they did not fill the application form for financial aid.12 I use a second dataset that contains program level information between 2008 and 2015. I use this sample to study the how past financial aid expansions affected enrollment of different groups of students. 11 I consider only one year in this sample because of current data restrictions. A future version will use data from 2011 to 2015. 12 Analysis of administrative data on parental income from another dataset shows that non-applicants to financial aid are similar to the wealthiest students who applied. This is consistent with the scenario where eligibility criteria for scholarships is public and known by many. 9 3 Enrollment Effects of Aid: Regression Discontinuity Estimates Let ti denote student i’s average Mathematics-Language admission score, and Ii her family income that are used to determine scholarship eligibility. Let t be the test score eligibility cutoff, and I the income eligibility cutoff. Conditional of Ii ≤ I, ri = ti − t (1) determines scholarship eligibility (ri ≥ 0).13 These eligibility rules are publicly known and publicized, but students have no control over their exact test score, creating a regression discontinuity design (RDD). This is the same strategy pioneered by Van der Klaauw (2002) and used in a growing literature by others (e.g. Goodman (2008), Cohodes and Goodman (2014)). In the Chilean context, Solis (2017) uses the same strategy to study the state guaranteed loan program. Figure 2a plots scholarship assignment as a function of ri among the income eligible population in the analysis sample, confirming the sharpness of scholarship assignment in this sample. Plotted points are conditional means for all applicants in a two-point bandwidth. The plots also show estimated conditional mean functions smoothed using a second order polynomial. Not everyone above the cutoff is awarded the scholarship, as only students enrolled at an eligible institution will receive it. The bottom panel shows the average enrollment rate at eligible institutions for different levels of the running variable. The jump in the average enrollment rate at the eligibility cutoff constitutes non-parametric evidence of the effect of scholarship eligibility. I construct non-parametric RD estimates for the effect of scholarship eligibility on enrollment using the average Mathematics and Language score as the running variable among students with eligible income levels. This initial set of results provides the basis of the price variation used in the demand estimation. Specifically, I estimate a local linear regression of the form: yi = γ0 (1 − Di )ri + γ1 Di ri + ρDi + i , (2) where the variable yi is an indicator for enrollment of student i, xi is the running variable centered at the discontinuity, Di is an indicator for xi ≥ 0, and the coefficient of interest is ρ. Equation 2 is estimated from a kernel-weighted least squares fit, using a triangular kernel centered around the scholarship eligibility threshold and narrowing the data used in estima13 Because I only observe income quintiles, I cannot exploit the discontinuity of eligibility on income. 10 tion to a data-driven bandwidth selected with the criteria used in Imbens and Kalyanaraman (2012). As it has been suggested by practitioners, I report results for multiple bandwidths and provide robustness of these results in Appendix B, including other bandwidth selection procedures and parametric specifications using higher order polynomials. None of these choices produced meaningful differences in the results. Table 4 shows that being eligible for a scholarship increases enrollment by 9.5 percentage points from a basis of 28% enrollment at eligible institutions. Columns (2) and (3) show the extensive and intensive margin effect of scholarship eligibility, respectively. Being eligible for a scholarship increases enrollment at any institution by 6.5 percentage points over a basis below the cutoff of 69%. Column (3) shows that enrollment at ineligible programs drops by 3.1 percentage points over a below-the-cutoff basis of 34%. This is direct evidence that students trade-off prices and other program characteristics when deciding their enrollment, and that some students will enroll at eligible institutions only if being awarded a scholarship. I show heterogeneity of these effects by estimating equation (2) for each of the three eligible income quintiles. Panel A in Table 5 shows that the effect of scholarship eligibility increases with income. Scholarship eligibility increases enrollment at eligible institutions by 7.7 percentage points (28.5% over the baseline rate) for students in the first income quintile, while it has an effect of 10.9 percentage points (37% over the baseline rate) for the third income quintile. This larger response from higher-income students mixes differential intensive and extensive margin responses. Panel B of Table 5 presents very similar responses on overall enrollment by income quintile, which are around 6.5 percentage points, and shows that as a proportion of the baseline, enrollments are very similar across quintiles (approximately 10%). The differences presented in panel A are partly due to the difference in substitution between vocational institutions and universities. Panel C shows that poorer students are less likely to substitute away from vocational/technical higher education institutions when offered a scholarship. This suggests that lower-income students are somewhat less responsive to price subsidies; however, a formal test of these differences does not reject the null of equal effects. Finally, I consider a placebo test using the fifth and richest income quintile. Students in this quintile are not eligible for a scholarship, so their enrollment decisions should not be affected by crossing any threshold associated to scholarship eligibility. Using the common cutoff for the first three income quintiles, the last column of Table 5 shows little evidence of an effect and only a marginally significant result for one of the three measures of enrollment. Appendix B reports results to address several possible threats to the validity of a causal interpretation of the results in Table 4 and 5. In particular, I look at the potential difference in demographic characteristics of students at both sides of the scholarship eligibility cutoff. 11 There are a couple of covariate contrasts that are significant, although very close to zero, and overall these gaps seem consistent with the idea that threshold crossing is indeed a good experiment. Together, these results strongly suggest that enrollment decisions change discontinuously at the cutoff of scholarship eligibility. The positive effect of scholarships on enrollment is consistent with the evidence of a large quasi-experimental literature.14 These results are best interpreted as the causal effect for individuals at the discontinuity. I will embed this quasi-experimental effect on a structural model and extrapolate to estimate preferences for the overall population imposing further restrictions that are discussed in section 5. 4 General Equilibrium Effects and Crowding Out The evidence above shows that students respond to reduction in tuition by changing their enrollment decisions. I now develop a simple model to explain how expansions of financial aid eligibility can create crowding out. This framework will also suggest the primitives that I need to estimate in order to measure crowding out. Then, I present direct evidence of crowding out using past financial aid expansions in a difference-in-differences strategy. 4.1 A Stylized Model This subsection presents a stylized model to provide the intuition for crowding out, the role of colleges’ admission cutoffs, and the objects quantified in my empirical strategy. My analysis builds on Azevedo and Leshno (2016), who show the role of admission cutoffs as prices in a general two-sided market with a continuum of agents on one side of the market, and on Agarwal (2017), who presents an empirical discussion of price and quantity policies in these types of markets. Consider a unit mass of students distributed along three dimensions: willingness to pay for the selective college (W ), income (I), and a scalar test score used for admission at the selective college (t). In this model everyone enrolls, and students decide between two options: one selective college and another non-selective college. The selective college has a binding capacity constraint that is initially fixed, while the non-selective college can admit all applicants. Willingness to pay at the non-selective college is normalized to zero, so that students with W ≥ 0 will prefer to attend the selective college. Applicants to the selective college are admitted based on their test scores, creating a cutoff admission rule. Financial aid to attend the selective college is initially targeted to students with income I ≤ 0.15 14 15 See Deming and Dynarski (2009) survey of this. Considering a subsidy only to the selective college mimics the Chilean scholarships studied before. It can 12 Figure 3a shows the group of applicants to the selective college in the space (I, W ), and in this model corresponds to everyone with W ≥ 0. A subsidy s to study at the selective college for students with I ≤ 0 implies that all students whose willingness to pay is W −s ≥ 0 apply to the selective option. Receiving a scholarship or making the current scholarship more generous compensates for a negative willingness to pay, attracting new applicants that have lower valuation for the selective college. Figure 3b shows in dark area the applicants that would be added in the case where a more generous scholarship is offered to everyone (free tuition). The selective college admits the most qualified applicants until reaching its capacity limit. After a tuition subsidy increases the number of applicants, if new applicants are more qualified than the baseline admitted students, the last admitted student would have a higher score, raising the admission cutoff. In that case, some of the students admitted at the baseline will be displaced by newcomers. Figure 4a shows the group of accepted students at the selective college in the space (I, t), with the normalization of the initial admission cutoff to be t = 0. A scholarship to a small fraction of students (of mass 0) will induce new applicants to enroll in the selective program if they have test scores above the initial admission cutoff. This is the effect captured by the regression discontinuity in subsection 3. However, in the presence of more generous and less targeted scholarships, the pool of applicants increases as Figure 3b shows. Figure 4a presents the case where new applicants have higher scores than the initial admission cutoff, in which case the admission cutoff needs to be raised because of capacity constraints, thereby displacing students in the red area. The importance of crowding out and who would be affected is determined by students’ preferences, test scores, and scholarship targeting. This model can be used to understand complementary policies to free tuition that could alleviate crowding out. In particular, I consider two policies that are often seen in practice: (i) Capacity expansion and (ii) Means-testing. In the model, the mass of students in the red area of Figure 4b represent the supply expansion needed to absorb all new applicants and keep the admission cutoff at the pre-free college level. This is a direct and simple way to deal with crowding out, but it might not be feasible if the required expansion is too large. If the objective was to keep enrollment rates by income level at least at baseline levels, without worrying about the identity of who is accepted at the selective college, then a smaller supply expansion would be necessary, but its magnitude will depend on the empirical distribution of preferences, test scores, and income. A more sophisticated means-testing of financial aid might also alleviate the effect of also be interpreted as a scholarship for a proportion of the tuition. As selective programs are more expensive this translates into a bigger subsidy to attend the selective option. 13 crowding out. To incorporate the objective of making the policy universal, I consider a linear schedule of financial aid that depends on income. Figure 5 shows different meanstesting policies and the corresponding set of new applicants of each policy. These policies reduce the set of new applicants with high-income by giving poorer students a larger subsidy. Whether these policies reduce crowding out or not depends on the mass of students with high test scores that each policy attracts to the selective college. Section 5 explains how to estimate the components of this model to quantify crowding out and who is affected. Next, I show direct evidence that crowding out has happened in past eligibility expansions. 4.2 Effects of Previous Expansions I estimate the effect that earlier scholarship eligibility expansions had on the enrollment of low-income students who were eligible for these benefits throughout the last decade. Figure 1 summarizes all the changes in eligibility for 4-year program scholarships. The first change happened after a period of constant eligibility criteria from 2007-2011, when only high-achieving students in the poorest 40% were eligible, and I look at expansions in the period from 2012-2015, when a new income quintile was included and students could access scholarships with half a standard deviation lower scores. I use a difference-in-differences strategy that exploits the changes in eligibility criteria over time and the differential exposure that programs had to the expansion. In particular, I approximate the exposure that each program had to these expansions using the average share of students in the poorest 40% between 2008 and 2011. This intends to proxy programs’ chances of receiving new scholarship beneficiaries after the introduction of financial aid expansions. I estimate the following equation, yjt = α + ρsharej · 1{t > 2011} + τt + τj + jt (3) where yjt is the outcome of interest for program j and year t, sharej is the average share of low-income students between 2008 and 2011 at program j, τj and τt are program and time fixed effects, and ρ is the parameter of interest that measures the effect that the post2011 financial aid expansions had on programs with a higher pre-2012 share of low-income students. Estimates of equation (3) show negative effects on enrollment for low-income students that were eligible for scholarships throughout 2007 and 2016. This can be seen in columns (1) to (3) of Table 6, where I present different specifications of equation (3) and the dependent variable is the share of students enrolled in program j coming from the poorest 40%. All 14 specifications are stable and suggest that programs with one standard deviation higher share of low-income students in the pre-expansion years experienced a reduction of approximately 5 percentage points (10% over baseline) in their share of low-income students. The results therefore reject the null of no effect of the size of the pool of eligible students on enrollment for low-income students who were eligible for scholarships before 2012. On the other hand, columns (4) to (6) of Table 6 show estimates of equation (3) using admission cutoffs as dependent variable. I find that programs with one standard deviation higher share of low-income students in the pre-expansion years had an increase of 0.3 standard deviations of test score in their admission cutoff post-eligibility-expansion. These results are in line with the prediction of the stylized model and with the hypothesis that financial aid expansions increase competition for programs. Finally, columns (7) to (9) present the same exercise using enrollment as outcome. The coefficients suggest that total enrollment at programs with a one standard deviation higher share of low-income students in the pre-expansion period results in a 10% reduction in total enrollment. This reduction in capacities might partly explain the increase in cutoffs and could be part of a strategy that programs followed to increase selectivity after 2011. Together these effects suggest that crowding out could be the outcome of financial aid expansions. Appendix C shows that the parallel trends assumption needed to causally interpret the difference-in-differences estimates is mostly met for the share of low-income students, but that some pre-trend exists for cutoffs and enrollment, and that therefore their causal interpretation might be problematic. These results are best interpreted as a test of the null hypothesis of no crowding out because the existence of displacement or spillovers complicates the causal interpretation of most quasi-experimental strategies. If a financial aid expansion pushes low-income students out of programs where competition increased, they might move to programs with lower demand. This means that the increased demand affects both treated and control programs, implying that non-randomized evaluation will overestimate the reduction in low-income students’ enrollment. Specifically, this scenario implies a violation of the Stable Unit Treatment Value Assumption (SUTVA).16 Nevertheless, in the absence of crowding out, the empirical strategy I used is a valid test for the existence of crowding out. This problem has been long recognized in the job placement assistance literature, where recent studies use clustered experimental variation to address this issue (Crepon, Duflo, Gurgand, Rathelot and Zamora, 2013). For this reason, and because I am interested in studying the design of universal financial aid policies, my main empirical strategy uses an equilibrium assumption and a local but large price variation to identify how free tuition would affect students’ demand for college. 16 Under SUTVA, potential outcomes for any student do not change with the treatment assigned to other individuals. 15 5 An Empirical Model of the College Market To quantify the consequences of a tuition-free policy on enrollment, I build on Fack et al. (2017) to create student-specific choice sets using realized admission cutoffs and estimate a random utility model where the price coefficient is identified by the local price variation induced by the regression discontinuity design of scholarship eligibility. This section describes this empirical strategy by first presenting the equilibrium notion that justifies using admission cutoffs to create individual choice sets. In subsection 5.2, I develop a general random utility model and present the parametric version I estimate. In subsection 5.3, I present an identification result that justifies the use of the regression discontinuity design to identify price coefficients in a random utility model. Finally, I present some of the estimation details, with further details provided in Appendix E. 5.1 Equilibrium I assume that matches between students and programs are pairwise stable, i.e. each student enrolls at her most preferred program for which she is eligible for admission. Formally, let i ∈ I = {1, 2, . . . , N } be the set of students to be matched with a finite set of programs J = {1, 2, . . . , J}. Each program j has a capacity cj , which is the number of students it is willing to admit. Each student i has a preference ordering i over the set J ∪{0}, where {0} denotes not enrolling in college, and a vector of scores ei ∈ [0, 1]J that describe programs’ 0 ordinal preferences for the student.17 That is, if eij > eij , then program j prefers student i to student i0 . Because in Chile programs’ preferences depend only on these weighted scores, I assume that all students are acceptable to programs. A match, given by the function µ : I → J , describes the allocation of students to programs. Let µ−1 (j) denote the set of students enrolled in program j. A pairwise stable match satisfies two properties for all student i ∈ I: 1. Individual Rationality: µ(i) 0. 0 2. No Blocking: if j i µ(i) then for all i0 ∈ µ−1 (j), eij ≥ eij . Individual rationality implies that there is no student enrolled in college who would rather not. This is a minimum requirement to interpret enrollment decisions as revealed preferences, and it accords with the voluntary nature of enrollment. The assumption that all students are acceptable to programs makes the allocation individually rational to programs as well. 17 In the Chilean context this corresponds to the weighted score using GPA and the national admission test scores. 16 The no blocking condition means that no student prefers a program (to her current match) that would admit her in place of the least qualified admitted student in the program if it is using all its capacity. This least qualified admitted student in each program j defines the admission cutoff at that program, Pj = min eij −1 i∈µ ∀j ∈ J , (4) (j) with Pj = 0 for programs that do not use all their capacity. Based on these cutoffs, student i is eligible for admission at program j if Pj ≤ eij and this defines her choice set, S(ei , P) = {j ∈ J |Pj ≤ eij } (5) Proposition 1 from Azevedo and Leshno (2016) for the continuum case highlights the relation between cutoffs and pairwise stability. Proposition 1. µ is stable if and only if µ(i) is student i’s favorite program among those in S(ei , P) for all i ∈ I. Therefore, a pairwise stable matching corresponds to each student choosing her favorite program conditional on being accepted at the vector of cutoffs P. The use of pairwise stability as an equilibrium assumption is appropriate for college admissions systems with low frictions. It is motivated in my case by the general admissions rules used in college admissions. Unlike American universities, Chilean colleges use formulaic aggregates of high-school test scores to determine admission. A threshold score for each college is published annually, and these scores are stable one year after the next. Students should therefore be aware of which programs are likely to admit them after their tests have been scored. Moreover, most selective universities use a centralized stable matching algorithm to further avoid instances where a student is not enrolled in her most preferred eligible option. A potential threat to this assumption is lack of knowledge about admission requirements at programs outside the centralized admission system. Appendix A.3 presents examples of information on the minimum test scores and grades required for admission at two universities inside and outside the centralized system. Similar information is given to students at fairs that high schools hold, and it can be easily accessed online.18 A second caveat is that strategic ranking behavior leads to ex-post regret on the assigned program. As explained 18 Hastings, Neilson and Zimmerman (2016), using a survey, report that 40% of students get information at these fairs. 17 in subsection 2.2, students are limited to ranking up to ten programs, and a minority of institutions imposes a lower cap on the maximum number they will consider. However, these restrictions do not seem to be binding as only 1.5% of students rank ten options, and 88% of students are accepted at one of their top three options. The stability assumption may be implausible in other college markets where admission criteria are not as transparent and publicly available, and where selective institutions do not hold a centralized assignment system that ensures violations do not occur. I use realized cutoffs for programs in the centralized system and the average MathematicsLanguage score of the least qualified enrolled student for the rest of the 4-year programs to construct student-specific choice sets. While the realized cutoffs are unpredictable for students (with cutoffs at the centralized system with a standard deviation of 0.18 standard deviations of the test), they are very stable from one year to the next, with a correlation of 0.95.19 Cutoffs from previous years are publicized by universities before students apply in the case of the centralized system, and they are usually established upfront for programs in the decentralized system. This creates a scenario where, up to the change in the realized cutoffs, students know their eligibility for admission before applying. Absent changes in the cutoffs from year to year, students would need to apply only to one option. Being able to rank multiple options helps students deal with cutoff uncertainty and ensures they are admitted at their preferred option once cutoffs are realized in the current year. The persistence of cutoffs at some programs makes students select their ranked alternatives, omitting options for which their scores are too far down from the previous year’s cutoff.20 Subsection 5.4 explains how I use this assumption to construct student-specific choice sets in my data. 5.2 Preferences I follow the discrete choice literature modeling the latent indirect utility that represents the preference ordering (i ) as a function of student observed characteristics, and programs’ observed and unobserved characteristics. The indirect utility student i gets from enrolling in program j is given by uij (xj , zi , wi , γi , ij , δj ; θ) = v(xj , zi , γi , ij ; θ) − π(wi )priceij + δj (6) δj = βx1j + ξj , 19 Regressing cutoffs using only last years’ cutoffs results on an R2 of 0.9986. Appendix A.4 presents graphical evidence of this behavior. Artemov, Che and He (2017) present evidence of this type of behavior in the Australian college admissions. 20 18 where xj is a vector of observed characteristics for program j and x1j is one of those characteristics, zi is a vector of observed student i’s characteristics, and wi are student characteristics other than her admission test scores. ξj is a program-specific unobserved characteristic, γi captures idiosyncratic tastes for program characteristics, ij contains idiosyncratic taste for program j, π(wi ) is a student-specific price responsiveness, and priceij is the tuition student i faces for program j, which is constructed as program j’s tuition minus the scholarship student i gets enrolling at program j. Scholarships are determined by student i’s average Mathematics-Language score ti and income eligibility. Conditional on a value of ti = t and on income eligibility, scholarships do not depend on the identity of students. I follow Berry et al. (1995, 2004) and include ξj as an unobserved (to the econometrician) program component that captures a quality characteristic observed by students. The fact that programs might set prices considering the desirability captured in this unobserved component implies that priceij 6⊥ ξj . This formulation allows for heterogeneous preferences, price endogeneity, and semi-additive preferences conditional on observables. Although quite general, equation (6) embeds four assumptions that I make throughout the paper. First, ξj is a scalar unobserved program characteristic that does not depend on the test score of student i, ti . This assumption is made in much of the existing empirical work, and it restricts the form of the endogeneity problem. Second, priceij , ξj , and a program characteristic, x1j , enter the indirect utility linearly. This assumption is commonly used in practice as well, and it implies that the program unobserved component and x1j affect the utility only through δj , and linearity in price facilitates the extrapolation of preferences using the limited quasi-experimental price variation in my data. Third, students’ indirect utility depends only on their own assignment and not on other students’. I make this assumption in order to focus on the causal effect of tuition, which is the main channel of interest in my counterfactual analyses. In this sense, this assumption does not limit my ability to include baseline peer composition characteristics, but it limits my analysis of the effect that changes in peer composition have on demand (coming from price changes). These first three assumptions are standard in the school choice literature (e.g. Hastings et al. (2006), Neilson (2013), Fack, Grenet and He (2017), Abdulkadiroglu, Agarwal and Pathak (2017)). Finally, the source of variation that allows me to identify price coefficients is based on the discontinuity of prices with respect to ti , and therefore I cannot identify heterogeneous price responses by test scores. This implies that π(wi ) does not include ti . However, students’ price responsiveness is in principle allowed to depend flexibly on other observable characteristics.21 21 I am working to get access to more years of data, which would allow me to incorporate test scores in the price sensitivity by using the change in eligibility cutoffs over years. 19 While my identification result does not make additional parametric assumptions, I use several assumptions to assist estimation in my sample. For estimation, I specify student i’s indirect utility for program j as uij = δj = K X k=1 K X sel αk zik xjk + γxj νi − π1 + π2 max{I, Iis } priceij + δj + ij (7) βk xjk + ξj k=1 ui0 = 0 where θ = {α, γ, π, ξ, β} are the coefficients to be estimated, and ij ∼ iid type I extreme value distribution. Iis corresponds to the simulated per capita income of student i drawn from the empirical distribution of income in the national survey (CASEN)22 conditional on student i’s income quintile. The functional form of income allows for non-linear responses, ensuring that low-income students’ utilities do not grow unbounded as income approaches zero. In practice, I use I to be 2 thousand dollars, which corresponds approximately to the 25th percentile of the income distribution.23 The normalization of ui0 = 0 is without loss of generality. In estimated specifications, zik contains an indicator of public high school, an indicator of voucher high school, average between Mathematics and Language admission test scores, an indicator of being in the 40% poorest part of the population (defined using family income quintiles), an indicator of having a mother with no high school education, and an indicator of having a mother with high school education. Additionally, I include an indicator for program j being in the same geographic region as student i.24 xj considers: the share of students enrolled in program j that belong to the poorest 40% of the population, the share of students enrolled in program j that graduated from a public high school, an indicator of j being a STEM program, the average Math-Language admission score among enrolled students, and the number of students enrolled in the program. xsel corresponds to j the average Math-Language score of admitted stud…
Purchase answer to see full attachment