The credits that count: how credit growth and financial aid affect college tuition and fees

Abstract

Using a two-stage least squares model, we build a macroeconomic model of supply and demand for US higher education as measured by enrollment. We find that college education benefits (e.g. relative earnings and employment level), credit factors (e.g. student loan amounts and household debt), and financial aid shift demand. Higher tuition prices increase the appeal of higher education for students but credit constraints put a barrier on demand growth. Tuition prices and debt levels are highly correlated, suggesting that students respond to higher tuition prices by borrowing. School's operating costs as well as tuition and non-tuition revenue drive supply. Schools can use tuition prices to signal quality, and relative demand-side price-in-elasticity allows them to raise prices. For the private institution sector alone, we see a higher level of consumer price sensitivity, with schools determining enrollment levels and adjusting tuition price accordingly.

Keywords:

• demand for schooling
• educational economics
• educational finance
• state and federal aid
• student financial aid

Acknowledgements

We would like to thank Charlie Brown, Christopher Best, Randolph B. Cohen, Brian Kim, Steven Kou, Klaus Ley, Tim Maull, Ed Rothman, Brian Rumao, Sampsa Samila, Tyler Shumway, Tuomo Vuolteenaho, and the seminar and conference participants at the National University of Singapore and the 2010 INFORMS Annual Meeting for their comments and guidance. All errors are our own.

Notes

1. Disposable income is included here because it is highly correlated with the other proxies for benefit of college, meaning its inclusion as an individual variable leads to cross-correlation and biased estimators in the final model.

2. Wealth is measured by endowment values and the gap between tuition revenue and expenditures.

3. Other proxy variables were considered in the original data set. These other variables were discarded in preliminary analysis as inferior proxies for underlying variables. The majority of data are from the Digest of Education Statistics maintained by the National Center for Education Statistics. A complete data set for the period 1976–2009 is constructed by extrapolation based on the available data points. The complete data set is provided in Appendix G.

4. It is possible that some sort of ‘ranking inflation’ exists, meaning that higher rankings overall may indirectly affect other metrics. This is not explicitly treated in our model. Relative metrics could be included in future studies at the individual school level.

5. We also conducted all analysis without the normalization, i.e. by modeling the effects of aggregate spending, cost, aid, and college benefits on total enrollment in terms of number of students. This model yields similar results and consistent coefficient signs, alleviating concern that our results are based solely on the chosen normalization.

6. Fox, John, Package ‘sem’, CRAN, 2010.

7. Note that since tuition falls out of the demand model, we use ordinary least squares with a first-order autocorrelation term to estimate the demand model.

8. Exceptions are benefit of college and credit effects per student in the demand model, which are both insignificant in the pre-test. Net tuition (per student) cannot be adequately pre-tested due to the presence of endogeneity, and because there is omitted variable bias in the pre-estimation model regressing enrollment on tuition price alone.

9. Note that net tuition price used in these models, though adjusted for institutional aid, is not adjusted for grants, meaning this variable does not necessarily represent the price actually paid by students. Grant aid is highly correlated with net tuition (correlation is 0.66), and therefore falls out of the final model. Creation of a grant-adjusted net tuition variable (i.e. tuition−institutional aid−grant aid) produces a model similar to that provided in Table 6.

10. Note that we exclude HEPI from the list of instrumental variables in these models, since our tuition variable is has either perfect or zero correlation with HEPI.

11. Note that endowment market value drives the non-tuition revenue principal component. Private school endowment data are based on Colby College's endowment, and public school data are based on Oklahoma State University. These schools were selected because their endowments fell at the median value in their category in 2009, out of the top 250 schools included in the NACUBO-Commonfund Study of Endowments (National Association of College and University Business Officers, 2009).

12. Wealth is measured by endowment values and the gap between tuition revenue and expenditures.

13. We might expect that this is not the case for for-profit institutions. However, we could not collect enough historical data on for-profit institutions to conduct this analysis.