The Effects of State Higher Education Policies and Institutions on Access by Economically Disadvantaged Students

Abstract

Do state government policies and institutions promote access to postsecondary education by economically disadvantaged students? I analyze the number of state residents receiving federal Pell grants relative to the college-age population raised in low-income households. Using data for 1993–2008, I estimate separate models for total Pell recipients and for those enrolled in public, private nonprofit, and proprietary institutions. I find consistent evidence that state spending on both need- and merit-based financial aid enhances access by economically disadvantaged students, with the effects of need-based aid being larger and more robust than those of aid based at least in part on merit. I also find that students in states with consolidated governing boards have slightly greater access to public and nonprofit institutions than those with statewide coordinating boards. Enrollment by economically disadvantaged students in the public and proprietary sectors also responds to the opportunity costs of attending college, as measured by the state unemployment rate. I do not find evidence of direct effects from state policies limiting affirmative action, or political elites’ ideology.

Appendix: Robustness Tests

Given the absence of perfect measures for many of the concepts in my models, I estimated a number of alternatives to see whether the findings hold up. The results are summarized in Table 7.

Table 7 Summary of robustness tests

First, I reestimated the models of state residents receiving Pell grants in Table 3 for the years 1999–2008 only. I did this because the lagged free or reduced price lunch (FRPL) rates for many states dropped sharply in the late 1990s. Given the 9-year lag, this means that the reported rates in these states were higher in the mid-1980s than the late 1980s. The reason for this is not clear. Harwell and Lebeau (2010) show a drop in the total number of lunches served around 1980 corresponding to a change in eligibility criteria, but that is a few years too early for the pattern I observe. In any event the results for 1999–2008 are largely the same as for 1992–2008. Some coefficients become only marginally significant with several hundred fewer cases.

Related to this, there are some observations where the value of the dependent variable is quite high, perhaps even implausible. For example, Table 2 shows that the maximum value of total Pell grant recipients as a percentage of the low-income cohort exceeds 100%. These cases are mostly in the mid-1990s, and many involve New Hampshire. I estimated my models omitting New Hampshire or, alternatively, all cases where total Pell recipients exceeds 80% of the low-income cohort. The substantive pattern of the results is the same, but the statistical significance of the financial aid variables and the unemployment rate increases.

An additional concern with the lagged FRPL rate is that there will be migration both into and out of each state during the 9-year lag. Molloy et al. (2011) find that the 5-year interstate migration rate for U.S. households during 1980–2000 was between nine and 10%, but their data on annual migration rates show that households including children were less likely to migrate across state lines than those without children, and adults with a high school education were less likely to migrate than those with a college degree.

Perhaps the most obvious alternative to the lagged FRPL rate for purposes of estimating a low-income cohort is the contemporaneous state poverty rate. The poverty rate is by definition lower since households up to 185% of the poverty rate qualify for FRPL. On the one hand, the poverty rate is not age-specific, and the ratio of potential students slightly above the poverty line to those below the poverty line varies across states and years. On the other hand, the poverty rate is based solely on household size and income and is not affected by the other eligibility criteria for FRPL that have been added over time (see Snyder and Musu-Gillette 2015). The correlation between the lagged FRPL rate and the contemporaneous poverty rate in my data set is 0.71.

When I substitute the contemporaneous poverty rate the coefficient on the state unemployment rate is positive and significant for proprietary institutions only. This, however, may be a statistical construct. The correlation between the contemporaneous poverty rate and the July unemployment rate is greater than between the lagged FRPL rate and the unemployment rate (0.51 vs. 0.39). This means there is a negative correlation between the unemployment rate and the dependent variable by construction that offsets the effect of opportunity costs.

The other piece of the cohort of potential college students is the population age 18–24. Of course, not everyone in this age cohort is actually eligible to attend college; some haven’t graduated from high school. If I substitute total high school graduates from the previous spring or the previous several springs, I lose some cases due to missing data on private high school graduates. Nonetheless, there are no substantive changes.

When I use the median sticker prices for in-state students reported in IPEDS rather than net tuition and fee revenue per student, the coefficient for 4-year public institutions is consistently positive while that for 2-year institutions is consistently negative in all models. However, results for my other independent variables are largely unchanged.

State spending on both need- and merit-based financial aid grants is very asymmetrically distributed, as can be seen from the descriptive statistics in Table 2. I added quadratic terms to allow for declining marginal effects but the z-score of the coefficients for the first-order terms remained 1.6 or above except for merit-based aid for proprietary institutions. Similarly, the results are largely unchanged if I simply omit Georgia and South Carolina, which have by far the most merit-based aid.