Testing Federal Student-Aid Fungibility in Two Competing Versions of Federalism



This study employs twenty-three years of federal, state, and institutional student-aid data to investigate the effect of federal grants to students on the behavior of higher education institutions. Unlike previous studies, this research differentiates between federal aid programs according to whether a federal program follows "redivision federalism" or "cooperative federalism" concepts. Pell grants, exemplifying the former, appear highly fungible and inversely related to institutional grants, while campus-based programs, exemplifying the latter, appear less fungible and positively associated with institutional grants. The results suggest that grants are more important to higher education opportunity than some previous research has indicated.


Federal student grants not only help financially needy students attend college, they also have an effect on the budget decisions of colleges and universities. The direction and dimensions of this effect, however, are in dispute.

The purpose of this study is to examine the effects of federal student-aid on institutional behavior specifically in the context of federalism, to explore how different federal student-grant programs fit different conceptions of federalism, to discover if institutions have adjusted their own institutional grants in response to different federal grant programs, to estimate the size of any such effects, and to review related research.

Pell grants are the primary focus in this research. The Pell program is the major effort of the federal government to assist financially needy students attend college with the help of grants and is the most disputed as to its effects. Other federal student-aid grant programs are a secondary focus.

RESEARCH BACKGROUND

Research in this area in the past several years is contradictory. Larry L. Leslie and Paul Brinkman (1988), summarizing research in a meta-analysis, found that total grant-aid is associated with increased higher education access for the financially needy. U.S. Department of Education (1995) evaluations, however, have been unable to demonstrate that Pell grants in particular, despite their domination in size compared to other grant programs, are associated with such increases. Thomas Kane (1994) concluded that "the evidence for the effectiveness of Pell grants in promoting enrollment by low-income youths is ambiguous." The Congressional Research Service (1996) has advised Congress that there is no consensus in the research literature on the impact federal student-aid programs have had on access.

Michael McPherson and Morton Schapiro (1991a) suggested that federal student financial aid represents fungible money in institutions' budgets. It is axiomatic among many budget professionals, but not widely acknowledged outside the profession, that institutions can and do use federal student-aid to reduce their own unrestricted institutional aid to students in order to meet other budget needs, such as faculty salaries or new programs. If true, this would provide an explanation why Leslie and Brinkman found that grants in total increase access to higher education, but why others have been unable to find a specific association of Pell grants with increased access. A characteristic of money in unrestricted fund accounts is its fungibility, that money from one source, such as Pell grants, can be offset by a change in another source, like institutional student-grants. McPherson and Schapiro suggested in theory that changes in Pell grants would be met with changes in the opposite direction in institutional student-aid, but their empirical research showed that increases in federal student-aid were met in certain cases with increases, not decreases, in institutional student-grant aid, leaving the fungibility hypothesis unsupported and the riddle unsolved.

The contradictions may result from shortcomings in the studies' designs. Kane subtracted imputed Pell grants from average tuition and room and board to try to determine net price impact on enrollment. This procedure neglected the effect of other aid at institutions' disposal, which is substantial in the subset of institutions used in his studies, and may have led to substantial underestimates of the true grant effect on higher education access. Others have used the same procedure and likewise found no Pell grant impact on enrollment. McPherson's and Schapiro's study was based on changes measured only between two years, 1978-1979 and 1985-1986. Had they picked other years, the results would have been different. Pell grants, for example, were 30 percent higher in 1979-1980 than in the previous year, in constant dollars. The time frame they chose was marked by the volatility of the implementation and subsequent repeal of the Middle Income Student Assistance Act, as well the repeal of Social Security benefits for students. Such factors make conclusions based only on two years untrustworthy. Finally, McPherson's and Schapiro's equations included many other independent variables which, given the nature of their multiple regression analysis, may have been misspecified for purposes of testing fungibility.

Researchers in higher education finance likely have neglected fungibility issues based on McPherson's and Schapiro's comment that when looking at the relationship between federal and institutional aid, "potential offsets are not important factors" (1991b,317). The comment was not accompanied by an explanation as to why their evidence did not fit their theory, only a statement that the issue deserves more systematic treatment. Their findings, they cautioned elsewhere (1991a,74), were only a "first attempt, using readily available data, to make sense of a complex set of relationships," and that "the results should be interpreted with caution."

THEORY AND METHOD

To give it a better test, the fungibility hypothesis needs to be reconsidered in the context of two different versions of federalism. Research in higher education often has not distinguished among federal programs, neglecting the possibility that different types of federal programs may have different effects on institutional budget behavior. Research in elementary and secondary education more commonly addresses such distinctions. Stephen Craig and Robert Inman (1982), studying the potential effect of President Reagan's New Federalism on local school district spending, found that "how federal aid is given--with constraints or without--has a big effect." Earlier, Martin Feldstein (1978) studied different versions of federalism in Title I of the Elementary and Secondary Education Act.

Federal grants to higher education students fall into two categories of federalism, each with a long tradition. One is cooperative federalism. In this version of federalism, different governments--national on the one hand, and state and local (including institutional) on the other--exercise their constitutionally derived powers concurrently, attempting to work together to address specific problems, such as higher education access for the financially needy.

Federal grants with matching requirements, maintenance-of-effort standards, and performance expectations are typical examples of cooperative federalism. The higher education tradition of cooperative federalism includes the Morrill Acts of the nineteenth century, the National Defense Education Act of 1958, the Higher Education Act of 1965, and the State Student Incentive Grant (SSIG) program of the Education Amendments of 1972. Current cooperative federalism student financial-aid programs are Perkins Loans, Federal Work-Study (FWS), Supplemental Education Opportunity grants (SEOG), and SSIG. Each of these programs has had, as a defining characteristic during part or all of its existence, matching or maintenance-of-effort requirements designed to bring the efforts of different governments together in a mixture to work toward higher education access for the financially needy.

Other programs fall under what can be called redivision federalism, sometimes described as "sort and swap," or as a return to "layer-cake" federalism. In this version, different governments attempt to sort out their responsibilities, opting not for an exercise of concurrent jurisdiction over an area of government concern, but for separate jurisdictions.

The best current example of redivision federalism in higher education is the Nixon administration's attempt in 1970 (led by domestic adviser Daniel Patrick Moynihan) to have the federal government take responsibility for higher education access for the financially needy. Richard Nixon, writing to Congress, suggested, "No qualified student who wants to go to college should be barred by lack of money. That has long been a great American goal: I propose that we achieve it now" (Congressional Record, 1970). In the Education Amendments of 1972, Congress followed through by enacting Basic Educational Opportunity Grants for students (now renamed Pell grants). Congress simultaneously retreated from federal government aid to colleges and universities, fundamentally leaving that field to the states. Pell grants to students are funded 100 percent from the federal treasury, with no matching or maintenance-of-effort requirements placed on state governments or institutions, public or private.

The method used here to test the fungibility hypothesis is a time-series, ordinary least squares, multiple regression analysis of all grants (institutional, state, and federal) awarded since fiscal year 1974 (Pell grants' first funding). The unit of analysis is the institution. If institutional grants, the dependent variable, are not affected by other grants in the analysis, there will be little support for grant fungibility as an explanation for the research dilemma. If, however, institutional grants move as a function of other grants, this will be evidence of fungibility. How much institutional grants move in response to different types of federal grants, and in what direction, will define the fungibility characteristics of the grants. Rather than looking at only two years, this study tests the relationships over twenty-three years.

There are two specific hypotheses: (1) that federal grants of the redivision federalism version will be characterized by negative regression coefficients, showing an inverse relationship with institutional grants and (2) that federal grants following cooperative federalism will show positive regression coefficients. To test significance, the results must have a one-in-twenty (.05) or less opportunity that the true regression coefficients are zero, the observations having been the result of random variation.

In the multiple regression equation Y = a + b1X1 + b2X2....+ bkXk + e, the dependent variable Y is aggregate institutional student grant aid in millions of dollars, with independent variables X1 through Xk in millions representing other sources of student grants. The equation's intercept, the value of institutional grants when other grants are zero, is a; the respective b's are coefficients of the independent variables; the error term e represents the residual or unexplained amount of institutional grants after accounting for the X variables. The constant dollar data are from The College Board (1995, 1996).
Rather than settling on one model when more than one model may explain institutional grant behavior, independent variables in this study appear in different combinations in various models, representing attempts to shape realistic combinations as they have presented themselves to institutions. The independent variables tested in this equation will include state grants, the "generally available" federal grant programs Pell, SEOG, and SSIG; "specially directed" federal aid programs (sizable in the 1970s--mostly Vietnam veterans' GI Bill grants and Social Security benefits), and other aid programs not purely grants but with various grant-like subsidy components, such as Perkins Loans, FWS, and federal guaranteed and direct student loans.

Some models employ fewer independent variables than others, leaving out the influence of loan and work programs. Some models combine independent variables, such as state grants and SSIG, because they often come to institutions from states in such a combination. Some models combine SEOG, Perkins, and FWS in the familiar "campus-based" aid package.

The models employ one of two additional variables to control for year-to-year changes, which, if left uncontrolled, could result in misleading conclusions. One such variable is full-time-equivalent enrollment (FTE); another is a linear trend variable, customary with time-series analysis, to account and control for linear trends of any nature, which would include the upward trend of enrollment increases as well as trends such as tuition increases and economic conditions.

To avoid hetereoskedasticity problems, in which unexplained amounts of the observations are correlated with an independent variable, showing bias, no models are considered in which there is an error term correlation with an independent variable. In recognition of potential autocorrelation problems, which can result in overestimating the effects of the independent variables, no models are considered that do not meet a minimum Durban-Watson value for the number of variables in the model and number of years analyzed. To minimize multicollinearity, which results in inadequate variability among independent variables to evaluate their effects on the dependent variable, no models are considered in which the coefficient of determination, R-squared, is greater for an independent variable, when regressed against the other independent variables, than is the model's coefficient of determination for the dependent variable itself.

DATA AND ANALYSIS

Table 1 lists the variables tested in various models. The table also lists each variable's median and interquartile (25th to 75th percentile) range to indicate relative size and dispersion.



Table 2 illustrates sixteen models, following an example model, that meet the standards set forth by the method. The models are numbered in the first column for identification and are ordered only by number of variables. The second column identifies the independent variables used in the model. The third column identifies the regression coefficient of the independent variable representing redivision federalism in the regression equation. In every case, this is the Pell grant program. (In two models, it is the Pell grant program adjusted to eliminate proprietary institutions, which rarely make institutional grants to students.)



To explain briefly the meaning of the regression coefficient, it is the amount of change in the dependent variable, institutional grants, per unit change in the independent variable, Pell grants, controlling for all other variables in the particular model equation. In the example model, the regression coefficient can be interpreted as showing that over the history of the Pell grant program, an increase of $1 million in Pell grants is associated with a decline of $937,000 in institutional grants, controlling for the effects of campus-based federal aid and linear trends.

The fourth column indicates the relative contribution the redivision federalism variable (Pell grants) makes in each model toward explaining the observed amount of the dependent variable, institutional grants. This is helpful so as not to assume erroneously that the values of the regression coefficients in a model are proportional to their effects on institutional grants. A variable with a high regression coefficient may have little effect, while a variable with a low coefficient may have a greater effect. Because there is no single, infallible way to make the determination of relative contribution (Aigner, 1971, 101) three methods are used and compared for agreement. The first method deletes each X independent variable separately from the model and compares the resulting coefficients of determination (R-squared values). The second compares the size of the t-statistic. The third compares the size of the standardized regression coefficient. If there is disagreement, the first method is used. In the example, Pell grants rank first in explanatory value over campus-based programs, even though their regression coefficient is smaller.

The fifth and sixth columns provide the regression coefficients and ranks for the independent variable in the equation that represents cooperative federalism. In many of the models this is SEOG, but in some cases it is a combination variable that includes other such programs. In the example, an increase of $1 million in campus-based programs is associated with a $2.381 million increase in institutional aid, but the strength of this cooperative federalism variable is second to Pell grants in explaining institutional aid.

The regression coefficients in the example model meet the .05 level of confidence using a t test. This does not mean, however, that it is a good model. Other tests indicate a correlation of the residuals with the independent variables, revealing bias, and excessive autocorrelation, risking overestimation of the size of the coefficients. No attempt is made in this study to employ error correction techniques to adjust the coefficients of any model in order to meet the methodological standards. Such techniques may mask a more serious problem of misspecification of the variables themselves (Kennedy, 1992,122). Moreover, this study's hypothesis involves testing for the effects of programs representing different versions of federalism; it is not an econometric analysis for purposes of forecasting.

Findings to Support the Hypotheses

The sixteen numbered models in Table 2 provide ample evidence to support the hypotheses of this research. Particularly noteworthy is that under all models meeting the methodological requirements, there is an inverse relationship between the redivision federalism program and institutional grants. In all models, this relationship meets the .05 level of confidence. Almost as noteworthy, in all of the fifteen models with cooperative federalism programs, there is a positive association of such variables with institutional grants, and seven of these meet the .05 confidence standard. These relationships hold whether controlling for the effects of state grants, federally guaranteed loans, specially directed federal aid, or other likely relevant variables entered individually or in combination. They hold in cases in which the control variable is FTE enrollment or the linear trend variable.

The rankings indicate that Pell grants are an important explanatory factor in determining institutional grants. In one of the models, Pell grants are the largest factor, and in the rest, the second largest.

The rankings of the cooperative federalism programs, conversely, are lower. In practical terms, this means that they are not as much of a factor in determining institutional grants.

The sizes of the regression coefficients for Pell grants range from .461 to .914. A cursory look at the b's suggests that historically, as Pell grants have gone up and down in the aggregate, institutional grants have moved by half as much or more in the opposite direction. This does not mean that a marginal change in Pell grants in the future would be met with a commensurate change in institutional grants in the opposite direction but, because of the high explanatory value rankings, this is a possibility to the extent that these variables still affect institutional decision making.

The sizes of the regression coefficients of various cooperative federalism programs meeting the .05 confidence level range from .722 to 5.733. Because of their consistently lower explanatory strength, however, changes in the funding of these programs are less likely to affect institutional grants than are funding changes in the Pell grant program.

Figure 1 illustrates the coefficients and the relationships of cooperative and redivision federalism programs to institutional grants.



A Closer Look at Two Models

A visual inspection of the results in Table 2 indicates two of the models are particularly worthy of more thorough investigation, which can be done in Table 3. This table shows, along with the regression coefficients, their standard errors, standardized coefficients, t-values, and P-values (the probability that the true regression coefficients are zero).

Model 15 is made up of grants only, replicating the grant environment in which institutional budget decisions are made. Model 16 includes all such grants plus loans and work-study funds, representing more closely the total student-aid funding environment.



In Model 15, the high standardized regression coefficient of the trend variable suggests that multicollinearity may explain the low t-value of SEOG (Welch and Comer, 1988, 269), but this is no reason to discard the model, inasmuch as it meets the methodological standards and because cooperative federalism variables often show the same lower t-values in other models as well. This is not a problem in Model 16, where there is adequate variability to result in impressive t-values for all the variables.

In Model 16, it should be pointed out, the cooperative federalism variable is combined with state grants on the basis that the intention of cooperative federalism was to provide incentives for increases in state grants as well as institutional grants. This combination avoids the multicollinearity problems present when all aid sources are entered individually into the regression equation. State grants are positively correlated with institutional grants in models where both appear.

The Stafford loan coefficient in Model 16 is negative, from which one might conclude that institutions consider the availability of Stafford loans and the willingness of students to borrow as a substitute for grants. The larger negative Pell coefficient, however, suggests that Stafford loans may only be a partial substitute during declines in Pell grants, as institutional grants apparently increase to compensate. This matches the conventional wisdom in the student financial aid community (and is confirmed in other models) that declines in recent years in Pell grants have been compensated for by a combination of Stafford loans and increased institutional grants.

Stability of the Models

Figure 2 illustrates how closely Model 16 fits the twenty-three years of observed institutional grants. The top chart shows two boundary lines between which there is 95 percent confidence that the true slope of the regression line is located. The R-squared figure shows that 98.1 percent of the variation of institutional grants is explained by the variables in the model.



The middle chart shows the more uncertain confidence boundaries for the time-frame McPherson and Schapiro studied. Had Model 16 been employed to explore the variance in these observations, only 66 percent could have been explained, modest for time-series data. Model 16's equation, tested over only seven years, would also fail a Durban-Watson test. On the other hand, there is little here to support McPherson's and Schapiro's retreat from their theory that institutional student-aid is affected by federal student-aid. It is obvious from visual inspection that the observations fit the pattern of the other years, and that the confidence boundaries encompass the other observations.

Model 16 is not stable over the entire twenty-three year period of observation, however. The bottom chart shows that prior to 1981, the relationship between institutional and other student-aid was unclear. The boundaries indicate no relationship; R-squared shows that the model for these years explains virtually none of the variance in institutional grants. A plausible explanation is that federal student-aid prior to 1981 was a mixture of need and non-need grants and loans scattered among different agencies, considerable sums of which may have been unknown to institutions. During the first two years of the Reagan administration, Congress terminated many students' Social Security benefits, other students were cut off from loan subsidies, and many Vietnam war higher education grant benefits ended, having run their course. Despite congressional promises, Pell grants were not expanded to take up the slack for low-income students among these former categories, putting upward pressure on institutional grants. This policy context has continued without substantial change over the ensuing one and one-half decades.

The effect of specially directed student-aid, such as veterans' benefits and Social Security, which had considerable effect during the pre-Reagan years, appears not to be decisive in this study of federalism, however. Four of the models tested omit the variable without appreciable difference in the outcomes. It is also apparent from visual inspection that the clustered values representing the pre-Reagan years fit into the longer term view.

Specification of Variables

No analysis would be complete without considering if are there other explanatory variables that should be considered in the models. McPherson and Schapiro included institutional revenues such as tuition income, return on endowments, and annual charitable giving. Because these are highly correlated, even in their constant dollar conversions, it is judicious to combine them into one variable to avoid multicollinearity.

Including in the models a constant dollar, combined variable representing these institutional revenues has a noticeable effect. For example, in Model 15, both the redivision federalism and cooperative federalism variables retain their sign, but lose their necessary level of confidence. In Model 16, the redivision and cooperative coefficients lose confidence and approach zero.

But in neither of these models does the institutional revenue variable itself meet the .05 level of confidence and it offers little additional explanation for the observations of institutional grants. There is yet another reason arguing against its inclusion: the underpinnings of multiple regression analysis itself. Models must be built on cause and effect relationships, with independent variables preceding dependent variables in time. At most institutions, annual giving and tuition are not revenues that must find a place to be spent; they are driven upward to help pay for increases in institutional student-aid, among other expenses, and therefore are inappropriate as independent variables. Although exceptions exist, most institutions have a difficult time providing for increases in institutional grants and their actions to pay for them come about as a result of trying to fund unmet student financial need. Including institutional revenue variables therefore fails both practical and theoretical tests for variable specification. This appears be a factor in McPherson and Shapiro's failure to find empirical evidence for their fungibility hypothesis.

The subject of revenues invites further questions about the effects of federal student-aid on tuition levels, however, because there remains a belief that increases in Pell grant funding lead to increases in tuition rates. This belief survives despite the fact that recent years have witnessed, in constant dollars, declines in Pell grants and increases in tuition rates. This study attempts only to control for tuition-rate trends in its investigation of other variables, but it is consistent with research (Hauptman with Merisotis, 1990, 67) showing that tuition is likely to be forced up to provide unrestricted revenues to offset declines in Pell grant appropriations, and would support the view that an increase in Pell grant funding would help hold down tuition increases.

With final regard to the specification of variables, this study finds that it is difficult to build a plausible model that shows a positive relationship between redivision federalism grants and institutional grants, or a negative relationship between cooperative federalism aid and institutional grants, using any of the variables in any combination. Even with questionable assumptions, such as omitting control variables, no such models met the standards of this study for rejection of the null hypothesis.

Type of Institution

The last analysis in this study deals with the question of aggregation. Should these data, collected from public and private, four-year and two-year institutions, be analyzed separately by type of institution to determine if they can properly be pooled for this study? The answer of course is yes, because public and private higher education institutions are inherently different in their financing and in their place in a federal system of government.

Unfortunately, few disaggregated data exist for either institutional grants or the independent variables. Nevertheless, a partially disaggregated set of data can be constructed for comparison of public and private institutions, if the analysis is limited to using Table 2's example model. The data represent twenty-one years of institutional and Pell grants, and nineteen years of campus-based programs, compared to twenty-three years for the aggregated data set used in the full models.

Despite the problems inherent in the Table 2 example model, by calculating partial correlation coefficients among the four variables, at least the direction of the relationships and their relative size can be compared. Table 4 shows the calculations in three partial correlation matrixes, first for public and private institutions together, followed by each treated separately.



Table 4 indicates that among both public and private institutions there is an inverse relationship between institutional and Pell grants, and a positive relationship between institutional grants and campus based programs, when controlling for all four variables, just as in the aggregates. The relationship is somewhat stronger between institutional and Pell grants at private institutions, and between institutional grants and campus-based programs at public institutions.

Comparing partial correlations does not provide a conclusive answer. A contrary view results from performing a Chow test on the respective regression coefficients, using the F statistic to determine if the data come from the same populations and can be pooled. But such a test is limited in this case due to the model's shortcomings described in the discussion of Table 2, and to the additional fact that the data are only partially disaggregated and do not adequately represent the characteristics of the populations. Included among public institutions in these data are two-year community colleges, low tuition institutions usually supported by local as well as state taxes, where institutional grants are less common. If the data could be disaggregated further to exclude community colleges from the data, the correlations and regressions comparing four-year public and private institutions likely would show less difference. Even with the community college data included, however, the partial correlations appear strong enough to suggest that federal student-aid programs affect institutional behavior similarly in both public and private institutions.

Public Policy
How institutional behavior is affected by different federal student-aid programs is a concern for public policy makers. Are student-aid funds appropriated by Congress offset at other levels or are they supplemented? What mechanisms of federalism are at work to account for these findings? Do the findings have import for future public policy decisions?

It is a premise of cooperative federalism that governments and other institutions working together can accomplish more than when each is left to go its own way. That appears to be confirmed here, based on twenty-three years of cooperative federalism data.

Looking back even further, to at least 1958 and the passage of the National Defense Education Act, federal, state, and institutional authorities have often worked together to try to provide increased higher education opportunity for those with financial need. But the relationship witnessed a hiatus in the 1970s with the establishment of Pell grants in the manner of redivision federalism, and resumed in the 1980s with more funding commitment from institutions than from federal or state governments.

One useful way of viewing the public policy relationship between institutional and Pell grants in the post-Pell grant period can be observed in Figure 3, which plots their arithmetic difference in millions of constant 1994 dollars by year.



In academic year 1973-1974, institutional grants exceeded Pell grants by approximately three billion dollars. Increases in Pell funding changed the relationship, so that by 1979-1980, Pell funding exceeded institutional grants by two billion dollars. Although the relationship between the two is not controlled for other variables, as it is in the regression models, one can nevertheless see in this chart the direction of the offsets. The raw data indicate that in four of the years, institutional grants declined in constant dollars.

In the 1980s and 1990s, the offsets moved the other way. By the end of 1995-1996, institutional grants in constant dollars were over four billion dollars more than Pell grants, illustrating a parabolic curve. On five occasions during these years, Pell grants declined from the previous year's levels in real terms.

Overall, Figure 3 indicates that Pell grants offset institutional grants nine times, and institutional grant increases offset Pell grant declines fourteen times.

This does not establish that offsets are necessarily bad. There is no suggestion here that institutions took improper advantage of federal aid fungibility characteristic of redivision federalism. Institutions that collectively reduced institutional aid during the years of Pell grant growth turned around and increased aid to help financially needy students during years of Pell grant decline.

On balance, federal taxpayers seem to have gotten the better of it, the federal government having first established a cooperative federalism relationship to get institutions committed to access for low income individuals, then switching to redivision federalism and holding to its promises for a few years, but then backing away and leaving institutions to pick up the tab. Figure 3 suggests that the pendulum has swung approximately a billion dollars beyond the previous high point of institutional commitment.

A danger exists should institutions now back away. Tuition increases, to some extent, are a means-tested tax on more affluent students to pay for Pell grant shortfalls for others. Hauptman and Merisotis (1990) estimated that twenty percent of tuition increases for the years they studied was attributable to raising revenues to pay for institutional student-aid. Tuition increases, however, may have reached their limits. Some colleges and universities have reduced tuition simply by dropping institutional financial aid, to fanfare. The federal government has probably played by redivision federalism rules as long as it can without institutions in large numbers dropping their commitment to access for the low income. If the next Congress pursues a higher education tuition price-containment agenda, institutional grants to the financially needy will likely be reduced.

There are no mechanisms now in place to keep institutional grants at their current levels, even though the cooperative federalism coefficients in this study may suggest otherwise. Two cautions must be expressed about the effectiveness of the maintenance-of-effort and matching mechanisms. First, the cooperative federalism programs have not always contained such requirements in their successive statutory reauthorizations. The requirements appear sporadically over the years, not consistently. Second, because the overwhelming majority of institutions that participate in the cooperative federalism programs also participate in Pell grants, one would expect that the cross-cutting requirements would result in similar coefficients regardless of program, but they do not.

One explanation for such observations is that the mechanisms as they were applied were not only desultory, but weak. Current-dollar maintenance-of-effort and relatively soft matching requirements for the cooperative federalism programs were not sufficient to have cross-cutting bite. The enforcement effort by the federal government, if any, is apparently lost to history. Institutions ran the programs in the tradition that they understood them; cooperative federalism programs were to be supplemental, the Pell grant program was to be a redivision of responsibility for basic access to higher education and therefore subject to offsetting.

The next Congress will consider increasing Pell grants and cooperative federalism programs as well as enacting President Clinton's proposed tuition tax deductions and credits. Based on this study, increases in Pell grants given current conditions will serve to redress the offset imbalance and will, to some extent, relieve pressures to increase tuition. In other words, an increase in Pell grants will aid some low income students, help many institutions with their budgets, and perhaps aid middle income families by holding the line on tuition.

As for cooperative federalism programs, although the higher regression coefficients might suggest higher leverage for such programs over institutional grants, in reality these programs are already considerably overmatched in the aggregate due to the continuing decline of the value of Pell grants and the commensurate increase in institutional grants.

It is difficult, based on this study, to predict the effect of any new favorable tax treatment of tuition expenses on tuition rates and student-aid. Fears that it could be offset by tuition increases and cuts in student-aid are reasonable given the history of the GI Bill effects on tuition after World War II (Finn, 1978) and the history of the Pell grant program. It would seem prudent to take two precautions in enacting any tuition tax deductions or credits: make clear that the intent is to supplement in the tradition of cooperative federalism, not to replace, current tuition subsidies and student-aid; and to establish stronger cross-cutting mechanisms in existing or redesigned programs to ensure that states and institutions do not offset the intended effects.

CONCLUSIONS

This study has explored two competing versions of federalism to explain seemingly contradictory research findings on the relationship between federal student-aid and higher education access. Empirical research to date has found that grants to students increase access, but several studies have been unable to find a specific relationship between access and the federal government's largest grant effort, Pell grants.

One explanation for this failure, advanced by McPherson and Schapiro, is that federal student-aid is fungible in institutions' budgets, and is offset by changes in institutional grants. Their empirical research, however, based on changes between two selected years, did not support the fungibility hypothesis.

This study suggests that the fungibility hypothesis can be supported with empirical evidence if it is understood that different federal programs have different fungibility characteristics. Cooperative federalism programs, characterized by matching and maintenance-of-effort requirements and designed to be supplemental to state and institutional efforts, are not as subject to offsetting actions as are redivision federalism programs, such as Pell grants, which have no such characteristics.

Research that seeks to find the relationship between federal student-aid and access, but fails to account for the fungibility of Pell grants, is not likely to add to an understanding of the relationship and is likely to underestimate the grant effect by the amount of the offsets. Research that employs the fungibility concept but does not distinguish between the two different versions of federalism at work in the mix of federal student-aid, runs a risk of underestimating the offsets and concluding that there are no offsets when in fact they may be considerable. Both of these traditions of research are therefore of limited utility in public policy analysis and formulation.

These conclusions are supported in this study by exploring sixteen fungibility models, each of which finds an inverse relationship between redivision federalism programs and institutional grants, and a positive relationship between cooperative federalism programs and institutional grants. The study uses twenty-three years of student financial aid data.

Unfortunately, data do not exist to explore the models further to determine how fungibility may vary by type of institution. Some evidence suggests that the dimensions of the coefficients may be different at public and private institutions, but does not lead to a conclusion that the competing versions of federalism must be defined by type of institution.

This study should be helpful in future research that attempts to discover more about the relationship between student-aid and higher education access. It may also help guide future data collection efforts by government agencies and other organizations.

This study should also provide useful information on which Congress can make future decisions on federal program authorizations, appropriations, and tax policy. The amount of student-aid that actually benefits students, as opposed to being netted-out by offsetting decisions at state or institutional levels, depends in part on the version of federalism favored in congressional decisions. This applies to programs on the revenue side of the federal budget as well as the expenditure side, inasmuch as tuition tax credits and deductions can be written either in the tradition of cooperative federalism, with its characteristics, or redivision federalism, which permits maximum fungibility of state and institutional funding.

It follows from this study that cooperative federalism programs, at the same level of federal support, will provide students more higher education opportunity than redivision federalism programs. The version of federalism chosen for each aid program is an important aspect of higher education opportunity in a federal government such as the United States.

REFERENCES

Aigner, D. (1971). Basic Econometrics . Englewood Cliffs, N.J.: Prentice-Hall.

College Board (1995). Trends in Student Aid: 1985 to 1995. New York: The College Board.

College Board (1996). Trends in Student Aid: 1986 to 1996. New York: The College Board.

Congressional Record (1970). Volume 116, Part 6:8018, March 19.

Craig, Steven G., and Robert P. Inman (1982). "Federal Aid and Public Education: An Empirical Look at the New Fiscal Federalism," The Review of Economics and Statistics, LXIV, 4, 541-551.

Feldstein, Martin (1978). "The Effect of a Differential Add-On Grant: Title I and Local Education Spending," The Journal of Human Resources, XIII, 4, 443-458.

Finn, Chester E., Jr. (1978). Scholars, Dollars, and Bureaucrats. Washington: The Brookings Institution.

Gillespie, D.A. (1983). Trends in Student Aid, 1963-1983. New York: The College Board.

Hauptman, Arthur, with Jamie Merisotis (1990). The College Tuition Spiral. Washington: The College Board and the American Council on Education.

Kane, Thomas J. (1994) "College Entry by Blacks since 1970: The Role of College Costs, Family Background, and the Returns to Education," Journal of Political Economy, vol.102, no.5.

Kennedy, Peter. (1992). Guide to Econometrics. Cambridge, Mass.: The MIT Press.

Leslie, Larry L., and Paul T. Brinkman (1988). The Economic Value of Higher Education. New York: Macmillan Publishing Company.

McPherson, Michael S. and Morton O. Schapiro (1991a). Keeping College Affordable. Washington, D.C: The Brookings Institution.

McPherson, Michael S., and Morton O. Schapiro (1991b). "Does Student Aid Affect College Enrollment? New Evidence on a Persistent Controversy," The American Economic Review. 81:1.

U.S. Department of Education (1995). Biennial Evaluation Report. Washington: U.S. Department of Education.

U.S. Library of Congress. Congressional Research Service (1996). The Higher Education Act: Reauthorization by the 105th Congress. CRS Report for Congress No. 96-806 EPW, by James B. Stedman and Wayne C. Riddle. Washington.

Welch, Susan, and Comer, John (1988). Quantitative Methods for Public Administration. Pacific Grove, Cal: Brooks/Cole Publishing Company.