Abstract
Public universities have pursued nonresident enrollment growth as a solution to the stagnation of state funding. Representatives of public universities often argue that nonresident tuition revenue is an important resource in efforts to finance access for resident students, whereas state policymakers are concerned that nonresident enrollment reduces opportunities for residents. This study investigated whether nonresident enrollment growth crowded out resident enrollment at public research universities using an instrumental variable identification strategy. For the sample of all public research universities, increased nonresident enrollment did not affect resident enrollment. For prestigious public research universities, nonresident enrollment growth had a negative effect on resident enrollment. The findings suggest that nonresident enrollment growth does not benefit resident access, as suggested by university administrators, nor does it harm resident access, as suggested by state policymakers. However, state policymakers may be concerned that nonresident enrollment crowds out resident access at prestigious public universities.
More recently, there is growing concern from state legislators (e.g., Koseff, 2016), policy think tanks (e.g., Haycock et al., 2010), and national media outlets (e.g., Strayer, 2016) that access to public flagship universities has become increasingly scarce for state residents. Nationally, from 2002–2003 to 2014–2015, the total number of resident freshmen enrolled in public nonresearch bachelor’s degree–granting universities increased 50%, whereas the number of resident freshmen enrolled in public research universities increased only 9%. During the same time period, nonresident freshman enrollment at public research universities increased 67%. Figure 1 illustrates these trends for the sample of public research universities utilized in this study. National media outlets, through vignettes of superlative high school students rejected from their state flagship university, have highlighted the perceived scarcity of resident access to public flagship universities and have placed blame on the increased emphasis on nonresident enrollment (e.g., Phillips & Belkin, 2014; Saul, 2016).
Average first-time freshman enrollment at public research universities.
The University of California provides a sharp example of this concern. From 2006–2007 to 2015–2016, resident freshman enrollment in the University of California System declined 2.7% from 33,530 to 32,630, whereas nonresident freshman enrollment increased nearly 400%, from 1,788 to 8,926 (University of California, 2016). The University of California System argued that nonresident enrollment growth is necessary to maintain quality and access for California residents:
Those funds subsidize the education of California students, especially as state funding has declined. More importantly, the increase in nonresident students has helped the University maintain its commitment to the California Master Plan by ensuring that every eligible California student receives an offer of admission. (Peacock, 2015, p. 1)
By contrast, California legislators have become concerned that nonresident enrollment growth undermines resident access to the state flagship university system. In 2015, after several years of funding cuts, the legislature offered the carrot of an additional US$25 million if the University of California System increased resident enrollment by 5,000 over 3 years (Assembly Bill 93, 2015). However, after fall 2015 freshman enrollment data showed a decline in resident enrollment and continued nonresident enrollment growth, Assembly Member Medina recommended stronger policy action: “Unfortunately, despite a strong directive and additional funding . . . additional statutory guidance is necessary to ensure all qualified California students have a fair chance at a world class University of California education” (as cited in Koseff, 2016, para. 3). As a result, a bill capping nonresident enrollment at 15.5% of undergraduate enrollment was proposed (Assembly Bill 1711, 2016).
Outside of California, some state systems and governing boards restrict nonresident enrollment or have considered changes to nonresident enrollment policies. For example, the University of North Carolina System can reduce the University of North Carolina–Chapel Hill’s campus-level operating budget if it exceeds a nonresident enrollment cap of 18% (University of North Carolina, 2014). In 2015, the Iowa Board of Regents approved a plan whereby 60% of state appropriations would depend on resident enrollment, thereby creating an incentive for Iowa public universities to increase resident enrollment (Burd, 2015a). By contrast, the University of Wisconsin Board of Regents approved a resolution that removed the cap on nonresident enrollment, so long as the University of Wisconsin–Madison enrolls at least 3,600 resident freshmen annually (Board of Regents of the University of Wisconsin System, 2015).
Although institutional leaders often argue that nonresident enrollment is necessary to finance access for residents amid state funding cuts (e.g., Blank, 2015; Peacock, 2015), many politicians and policy advocates argue that nonresident enrollment growth crowds out access for state residents (e.g., Haycock et al., 2010; Koseff, 2016). The debate even spilled into the 2016 U.S. presidential election, with Hilary Clinton endorsing state legislation that limits nonresident enrollment:
We have got to get back to using public colleges and university for what they were intended. . . . If it is in California, for the children in California. If it is in New York, for the children in New York. (Jaschik, 2016, para. 2)
However, empirical research has not examined the causal effect of nonresident enrollment on resident enrollment. Unfortunately, this leads politicians and policymakers to propose and debate policies about nonresident enrollment restrictions based solely on descriptive and anecdotal evidence.
The goal of this article is to provide an evidentiary basis for state policy debates by analyzing the effect of nonresident enrollment on resident enrollment at state flagship universities. Nonresident enrollment may be endogenous because enrollment management strategies for resident and nonresident enrollment are likely jointly determined. Therefore, we utilized an instrumental variable estimation strategy designed to isolate exogenous variation in nonresident enrollment demand. Specifically, we used state merit-based aid expenditure in other states as an instrument for nonresident enrollment, as prior research has shown that state merit-based aid reduced out-migration (Orsuwan & Heck, 2009; Zhang & Ness, 2010). For the sample of all public research universities, we found no relationship between nonresident enrollment and resident enrollment. However, for prestigious public research universities, nonresident enrollment growth had a negative effect on resident enrollment.
Literature Review
To provide a background for our analysis, we review empirical research on the determinants of resident enrollment, our dependent variable. We first focused on state policy determinants that influence resident enrollment. Second, we explored literature on institution-level factors related to resident enrollment.
State Policies That Influence Access to Higher Education for Residents
State Appropriations and Tuition Price
Winston’s (1999) statement that tuition price equals instructional cost minus subsidy is helpful for explaining the relationship between state appropriations, tuition price, and resident access. At public institutions, this subsidy is historically composed of state appropriations. Winston’s formula suggests that when state appropriations decline relative to enrollment demand institutions must raise tuition price, decrease costs (i.e., spending per student), or enroll fewer students.
Drawing from Winston (1999), state appropriations are expected to affect resident access through changes to the demand for public higher education and through its effect on resident tuition price. State appropriations to institutional general operating budgets subsidize the costs of educating students (e.g., instruction, student services, and facility maintenance). Therefore, when state appropriations increase, institutions have greater capacity to educate more students, a relationship found by Toutkoushian and Hillman (2012). Alternatively, growth in the college-going population relative to state appropriations has been found to be negatively related to resident college completion due to the decline in enrollment opportunities and lower spending per student (Bound & Turner, 2007).
Changes in state appropriations also affect resident enrollment through changes in tuition charged to resident students. A robust empirical literature found a negative relationship between state appropriations and resident tuition price (e.g., Koshal & Koshal, 2000; Rizzo & Ehrenberg, 2004). In turn, research generally found a negative relationship between tuition price and resident enrollment at public universities (e.g., Heller, 1997; Hemelt & Marcotte, 2011). This was particularly true for low-income and underserved minority students (Heller, 1999; Paulsen & St. John, 2002; Perna, Steele, Woda, & Hibbert, 2005).
More recent research has examined the relationship between tuition deregulation and access to higher education. In most states, policymakers at various levels above the university have authority to set price ceilings on resident tuition price (McBain, 2010). However, several states (e.g., Texas, Virginia, and Florida) have deregulated tuition-setting authority. Tuition deregulation in Texas caused tuition price to increase, particularly at selective public institutions (Kim & Stange, 2016), and was negatively related to enrollment at public research universities for Hispanic students, with mixed results for other groups (Flores & Shepherd, 2014).
State Financial Aid
State grant aid programs also subsidize enrollment for state residents. A large empirical literature found that state merit-based aid programs were positively related to enrollment at in-state public universities (e.g., Cornwell, Mustard, & Sridhar, 2006; Singell, Waddell, & Curs, 2006; Zhang, Hu, & Pu, 2016). In addition, research found that merit-based aid programs reduced out-migration by state residents to postsecondary institutions in other states (Orsuwan & Heck, 2009; Tout-koushian & Hillman, 2012; Zhang & Ness, 2010).
State need-based aid programs subsidize enrollment for low-income residents. Research generally found that state need-based aid programs increased access for resident students, particularly increasing the probability of attending public 4-year institutions (e.g., Castleman & Long, 2013; Toutkoushian, Hossler, DesJardins, McCall, & Canche, 2015). However, less is known about the effect of need-based aid on out-migration.
Institutional Policies Related to Access to Higher Education for State Residents
Although state policies can affect enrollment, this article focuses on institutional behaviors that affect enrollment for state residents. Institutions affect enrollment through “enrollment management,” which Hossler and Bean (1990) define as “efforts to influence the characteristics and the size of enrolled student bodies” (p. xiv). Generally, enrollment management offices control and integrate the activities of marketing, recruitment, admissions, and financial aid (Kraatz, Ventresca, & Deng, 2010). Although scholarship on marketing and recruitment is scarce, robust literatures on admissions and institution-based financial aid exist.
Admissions
Research on admissions policy found that college entrance exams have become more important in admissions decisions at selective public and private colleges (Alon & Tienda, 2007). In turn, this transformation in admissions policy negatively affected access for low-income students, who tend to score lower on entrance exams compared with affluent students (Alon, 2009; Bastedo & Jaquette, 2011).
A small empirical literature suggests that public universities set higher admissions requirements for nonresident applicants than resident applicants. Groen and White (2004) found that SAT scores for the marginal nonresident student at four selective public universities were 84 points higher than those of the marginal resident student. Using nationally representative data from the high school senior class of 2004, Cummings (2015) found that being a resident applicant increased the probability of admission by about 4.5%. However, she also found a negative interaction effect between residency and institution-level state appropriations, suggesting that admissions preferences for state residents erode when state appropriations are low.
Research has shown that institutions alter their admissions preferences as external conditions change. Winters (2012) found that growth in resident enrollment demand—operationalized as the state-level population of 18-year-olds—was negatively related to nonresident enrollment. Jaquette and Curs (2015) found a negative relationship between state appropriations and nonresident enrollment, suggesting universities were replacing lost state funding with nonresident tuition revenue.
Institutional Aid
Institutions can also affect access by manipulating tuition price and institutional aid. A large empirical literature analyzed the effect of institutional aid on enrollment decisions (e.g., Curs & Singell, 2010; DesJardins, Ahlburg, & McCall, 2006; Hurwitz, 2012; Monks, 2009). Although most studies cannot cleanly differentiate need-based versus merit-based institutional aid (but see Monks, 2009), the literature generally found that admitted students were quite sensitive to the size of institutional financial aid packages. This was true for high-achieving students (Avery & Hoxby, 2004) and high-income students (Singell & Stone, 2002), and Hurwitz (2012) provided evidence that low-income students were particularly sensitive to grant aid. A related phenomenon is the adoption of “no-loan” tuition policies; however, evaluations generally found that no-loan policies did not substantially increase access for low-income students (Hillman, 2013; Waddell & Singell, 2011).
The past decade has witnessed dramatic changes in expenditure on institutional aid and the focus of institutional aid. Total spending on institutional grant aid increased 86% from 2004–2005 to 2014–2015 (College Board, 2015b). At public doctoral-granting institutions, institutional grant aid for freshmen increased from US$1,430 in 2002–2003 to US$3,050 in 2012–2013 (in 2012 real dollars) (College Board, 2015b). Whereas institutional aid was historically used to increase access for low-income students (Ehrenberg, 2000), Doyle (2010) showed that institutional expenditure on merit-based aid increased dramatically over the 1990s and 2000s relative to expenditure on need-based aid. This result suggests that universities were increasingly allocating institutional aid toward the pursuit of academic quality. More recently, as declining state appropriations compelled public universities to become more tuition dependent, maximizing net tuition revenue has become a growing focus of institutional aid policy (Bosshardt, Lichtenstein, Palumbo, & Zaporowski, 2010).
A growing number of public universities have developed institutional aid programs that specifically target nonresident students (e.g., Burd, 2015b; DesJardins, 2001; Leeds & DesJardins, 2015). These policies are motivated by the goal of increasing net tuition revenue, as nonresident tuition price is typically more than twice that of resident tuition price, and/or the goal of increasing academic profile, as nonresident students often score higher than residents on college entrance exams (Jaquette, Curs, & Posselt, 2016). While aid offers increase with student academic achievement, grant aid is increasingly offered to both moderate-achieving and high-achieving students (Burd, 2015b).
To summarize, access to public flagship universities for meritorious state residents has been a valued goal for state policymakers (e.g., Koseff, 2016) but a growing concern for parents and access advocates (e.g., Strayer, 2016). Research on access to public universities tended to focus on state policies that potentially affect tuition price, state appropriations, or state financial aid. However, institutional enrollment management policy also affects access for state residents. Research on institutional aid showed a shift from need- to merit-based aid and a growing emphasis on using aid to maximize net tuition revenue. Although previous research suggested that public universities practice admissions preferences for resident applicants, this finding was based on old data. A growing number of public universities have adopted institutional aid programs for nonresident students, and nonresident enrollment has grown dramatically at public flagships over the past decade. In theory, nonresident tuition revenue could subsidize resident access—a variation of the high-tuition high-aid model (Curs & Singell, 2010; Turner, 2006). However, policymakers are concerned that nonresident enrollment growth crowds out opportunities for resident access. This concern has not been tested empirically. Therefore, this article analyzes the effect of nonresident enrollment on resident enrollment.
Conceptual Framework
This section describes potential mechanisms that could explain the relationship between nonresident and resident enrollment. Drawing from rational choice theory, we assume that actors (e.g., people, organizations) attempt to maximize utility, they have idiosyncratic preferences about which goals are important, and they face constraints (e.g., limited resources, regulations) to the realization of these goals. In the field of enrollment management, rational choice theory is the basis for the “iron triangle” of enrollment management (Cheslock & Kroc, 2012; DesJardins & Toutkoushian, 2005). Within this framework, universities pursue three broad enrollment goals: the access goal, which encompasses access for low-income students, underrepresented minority students, and—at public universities—state residents; the academic profile goal of enrolling high-achieving students to garner academic prestige; and the revenue goal of enrolling students that generate tuition revenue. The priority placed on each of these three goals differs across universities and changes over time. Although synergies can exist between the pursuits of different enrollment goals, the reality of scarce resources suggests that allocating resources toward the pursuit of one goal involves tradeoffs for other goals. For example, ceteris paribus, increasing expenditure on need-based aid for low-income students reduces expenditure on merit-based aid for high-achieving students.
Capacity constraints are central to the relationship between nonresident and resident enrollment. Capacity constraints arise when short-term enrollment supply is perfectly inelastic, implying that the university will not enroll more students even when student demand increases. Conceptually, for the purpose of this article, capacity constraints occur when a university caps enrollment within a given year (DesJardins & Bell, 2006). Said differently, the capacity constrained university may increase enrollment in the long run (across multiple years) but enrollment in the short run (within a given year) is restricted.
Capacity constraints may arise at public universities for several reasons. First, universities may lack the necessary physical capital (e.g., residence halls, classrooms) or labor (e.g., faculty, student service professionals) to enroll additional students. Second, Winston (1999) argues that the academic prestige of a university is substantially determined by the academic achievement of its student body. Therefore, universities can pursue prestige by fixing enrollment supply below student demand and selecting students with the highest academic achievement. Third, when revenue declines, universities may decide to restrain enrollment growth to retain a desired level of resources per student (Bound & Turner, 2007).
For capacity constrained institutions, we expect that nonresident enrollment growth crowds out resident enrollment. Said differently, if total freshman enrollment is fixed in a given year, and the university enrolls more nonresident students, then resident enrollment will necessarily decline. This remains true whether the capacity constraint is due to physical space (e.g., limited classrooms, dorms) or the desire to increase prestige.
For universities that do not face capacity constraints, organizational preferences for different enrollment goals render the expected relationship between nonresident and resident enrollment to be less clear. Given that net tuition revenue for nonresident students far exceeds that of resident students (College Board, 2015a; Jaquette et al., 2016), nonresident enrollment growth clearly contributes to the tuition revenue goal. In turn, universities may allocate nonresident tuition revenue toward either the goal of access for resident students or the goal of increased academic profile.
For noncapacity constrained organizations that prioritize the goal of access for residents, we expect a positive relationship between nonresident enrollment and resident enrollment because the university is likely to use tuition revenue from nonresident enrollment growth to finance access for resident students. Indeed, financing resident access has often been a stated rationale for nonresident enrollment growth by universities (e.g., Peacock, 2015). For example, public university operating budgets—which cover the cost of educating undergraduates—are mostly composed of tuition revenue and state appropriations (Weisbrod, Ballou, & Asch, 2008). Nonresident tuition revenue may be allocated to the general operating budget to satisfy resident enrollment demand, particularly when state appropriations decline. As another example, universities may use nonresident tuition revenue to expand need-based aid expenditure which is likely to disproportionally benefit resident students as low-income students are more likely to be residents than nonresidents.
For noncapacity constrained universities that prioritize the academic profile goal over the goal of resident access, nonresident enrollment is expected to have an ambiguous relationship with resident enrollment. For these universities, nonresident tuition revenue allocated to the general budget is less likely to be expended on resident enrollment growth and more likely to be expended on increasing spending per student (e.g., smaller classes, more extracurricular activities) as a means of attracting high-achieving students (Winston, 1999). The allure of nonresident students for universities concerned with academic profile is that they generate tuition revenue that can be used to attract students with higher test scores, which are often more likely to be nonresident students themselves. However, as competition for nonresident students intensifies, institutions may need to devote more recruiting, admissions, financial aid, and quality of life resources toward nonresident markets (Jacob, McCall, & Stange, 2013, Long, 2004).
To summarize, for noncapacity constrained universities, we predict a positive relationship for universities that prioritize resident access and an ambiguous relationship for universities that prioritize academic profile. Thus, for the majority of institutions, the relationship between nonresident and resident enrollment is an empirical question, as we are unable to make a definitive prediction of the direction of the relationship. However, we expect a negative relationship between nonresident enrollment and resident enrollment for institutions likely to have short-term capacity constraints, in particular high-prestige universities that are reticent to grow total enrollment. In the next section, we describe our empirical strategy to identify the relationship between nonresident and resident enrollment.
Research Design
Empirical Framework
In this study, we sought to identify the effect of nonresident enrollment growth on resident enrollment. Equation 1 shows a general institution-specific linear panel model, where Residentit is a measure of resident enrollment for university i in time t; Nonresidentit is nonresident enrollment, with β as is its associated population coefficient;
Two key assumptions must be satisfied to interpret the coefficient estimate on nonresident enrollment as a causal effect (Wooldridge, 2002). First, after controlling for covariates, there should be no relationship between nonresident enrollment and the unit-varying, time-invariant error component (ai). Therefore, to control for unobserved time-invariant heterogeneity, we utilize an institution-level fixed effects estimator that eliminated the potential correlation between Nonresidentit and ai.
Second, after controlling for covariates, there should be no relationship between nonresident enrollment and the unit-varying, time-varying error component (eit). Given that fixed effects estimators satisfy the first assumption, we were primarily concerned with bias due to violations of the second assumption that occur when within-institution variation in nonresident enrollment is correlated with within-institution variation in the error term. Bias in
In observational studies, attempts to control for all sources of bias through the inclusion of covariates are usually unsuccessful (Angrist & Pischke, 2009; Murnane & Willett, 2011). Furthermore, scholarship on enrollment management suggests that strategies related to resident and nonresident enrollment are jointly determined (Cheslock & Kroc, 2012; Hossler & Bontrager, 2014). Therefore, it is preferable to identify a source of variation in nonresident enrollment that is exogenous. This article presents an attempt to isolate exogenous variation in nonresident enrollment using an instrumental variables approach.
Instrumental Variables Identification Strategy
The instrumental variables estimation strategy calculates a consistent estimate of the population parameter, β, by using an instrumental variable, Zit, to isolate exogenous variation in Nonresidentit. The logic is that variation in Zit affects Nonresidentit, which in turn affects Residentit. We applied a two-stage least squares approach to estimate the instrumental variables framework. The first-stage equation (Equation 2) models the effect of the instrument, Zit, on the endogenous regressor, Nonresidentit, controlling for covariates,
The outcome equation (Equation 3) models the effect of Nonresidentit on Residentit, controlling for the same
Angrist and Pischke (2009) describe four assumptions necessary for the instrumental variables framework to estimate a consistent causal effect with heterogeneous potential outcomes. First, the relevance assumption states that the instrument affects the endogenous regressor. Second, the independence assumption states that the instrument is as good as randomly assigned with respect to the outcome and potential treatment assignments. Third, the exclusion restriction assumption states that the instrument only affects the outcome through the endogenous regressor. Fourth, the monotonicity assumption states that, for all units in which the instrument affects the endogenous regressor, the instrument has the same directional effect on the endogenous regressor. After describing the data in the next section, we describe our candidate instrument and discuss the instrumental variables assumptions in greater detail.
Data and Variables
Data
We created an institution-level panel dataset, incorporating institution-level data from the Integrated Postsecondary Education Data System (IPEDS) and state-level data from various standard sources.
Analytical Sample
The analytical sample consisted of all public 4-year institutions defined as research-extensive or research-intensive by the 2000 Carnegie Classification. We restricted the sample to research universities as prior research has shown that they were the most responsive in enrolling nonresident students due to changing fiscal conditions (Jaquette & Curs, 2015). The analysis period consisted of the academic years 1992–1993 through 2013–2014, though some sensitivity analyses utilized a shorter period. The sample started in 1992–1993 due to the availability of the migration component of the IPEDS fall enrollment survey and ended in the 2013–2014 academic year due to the availability of the instrumental variable.
Institution-year observations with missing values for resident enrollment and nonresident enrollment were dropped from analyses and not imputed. Missing institution-level covariates were imputed using the average of the within-panel 1-year lag and lead observations. All variables have been log transformed to reduce heteroskedasticity due to large variation in the size and scope of higher education institutions. Descriptive statistics for institution-level variables (prior to the log transformation) are presented in Table 1 for the analytical sample of 2,669 institution-year observations, which includes 18 years of data and 159 public 4-year research universities.
Descriptive Statistics
Note. Sample means are reported with standard deviations in parentheses. FTE = full-time equivalent.
Dependent Variable
Resident freshman enrollment was collected from the Residence and Migration subcomponent of the IPEDS Fall Enrollment survey. These data identify the number of freshmen enrolled in institutions of higher education at the fall census date from each state, U.S. territory, and those migrating from a foreign country. We defined resident freshman enrollment as any student whose state of residence was the same as the state in which the institution is located.
Prior to 2000–2001, the IPEDS Resident and Migration survey subcomponent was collected in odd academic years (e.g., the 1992–1993 academic year). Starting in 2001–2002, institutions could voluntarily submit this subcom-ponent in even academic years. Nonmissing observations from voluntary years were included in the primary analytical sample. Sensitivity analyses which excluded observations from voluntary years are discussed in the robustness portion of the “Results” section.
Independent Variable
The independent variable was nonresident freshman enrollment. This measure was defined as the number of freshmen whose state of residence differed from the state which the institution is located and includes students from U.S. territories and students migrating from a foreign country. These data were also collected from the Residence and Migration subcomponent of the IPEDS Fall Enrollment survey.
Control Variables
The choice of control variables was based upon two rationales. First, to increase precision and reduce omitted variables bias, we included time-varying covariates that plausibly affected resident enrollment and were correlated with nonresident enrollment. Thus, we included factors related to institutional demand for both resident and nonresident students. Second, we included variables to minimize threats to the independence and exclusion assumptions of the instrumental variables procedure.
At the institution level, we included tuition and fees for resident students, tuition and fees for nonresident students, and average institutional grants. To capture variation in institutional quality and resources, we included expenditures per full-time equivalent (FTE) student for the following categories: instruction, research, public service, services (academic, student, and institutional support), and auxiliary enterprises. To control for changing higher education conditions in other states, we included geographically weighted measures of nonresident tuition and public research university enrollment capacity (the ratio of the population of 18-year-olds relative to resident public research university enrollment). Considering the factors that could be correlated with resident enrollment at a particular institution and state-level merit-based aid generosity in other states, we controlled for the following state-level economic indicators: per capita income, annual unemployment rate, and total state population by the following age ranges: 12 to 17, 18 to 24, and 25 to 44. Finally, we controlled for state expenditures on needs-based and merit-based aid.
Instrumental Variable
To make our explanation concrete, imagine that we are trying to estimate the effect of nonresident enrollment on resident enrollment at the University of Alabama. We searched for instruments likely to affect demand for the University of Alabama of students from outside of Alabama but unrelated to demand for the University of Alabama by Alabama residents. To construct our instrument, we exploited state geographical boundaries and the fact that state financial aid policy only benefits students who resided within a state prior to their choice of higher education institutions.
State merit-based aid expenditure in other states served as the instrument for nonresident enrollment. An increase in state merit-based aid generosity decreases the relative cost of attending college in-state, thus increasing the likelihood that a student attends an in-state institution. A robust literature consistently found that state merit-based aid decreased out-migration of students for higher education (e.g., Orsuwan & Heck, 2009; Zhang & Ness, 2010). In addition, the vast majority of state merit-based aid programs restrict program eligibility to state residents who enroll at an in-state institution. Therefore, we argue that state merit-based aid generosity in a particular state (e.g., Georgia) is unrelated to the decision of residents from another state (e.g., Alabama) to attend an institution in their own state (e.g., the University of Alabama).
Using the same geographic boundary argument, we explored other potential instruments. State need-based aid generosity in other states and the population of 18-year-olds in other states were the two most promising alternative instruments. Unfortunately, neither of these instruments was strongly related to nonresident enrollment. Therefore, we did not include these instruments in our primary analyses but discuss their inclusion in the “Robustness to Alternative Specifications” subsection of the presentation of results.
Construction of the Instrument
The instrumental variable was defined, with respect to a focal university, as the weighted average of state-level merit-based aid expenditure per 18- to 24-year-old in other states. We constructed this variable using a gravity model approach often used in interstate migration research (Alm & Winters, 2009; Cooke & Boyle, 2011). This approach assumes that migration from a sending state (e.g., Georgia) to a focal institution (e.g., University of Alabama) was positively related to the population in the sending state. In addition, migration was assumed to be negatively related to the distance between the focal university and the sending state, with distance defined as the spherical distance from the focal university to the population centroid of the sending state.
For each unique combination of focal institution i, sending state s (excluding the state of the focal institution), and academic year t, we constructed a time-varying weight (wist) that assigned higher weights to states with larger populations (Popst) and states which are close to the focal institution (Distanceis). Specifically, wist is defined as follows:
For each institution i in each year t, the sum of the weights across all states equals 1. Alternative specifications in which the weighting scheme was allowed to decay more rapidly (by squaring distance) or more slowly (by taking the square root of distance) generally produced weaker instruments but did not qualitatively alter the primary results (findings are discussed in the “Robustness to Alternative Specifications” section).
After calculating the weights for each intuition-state-year combination, we constructed a weighted average of state merit-based aid generosity in other states (Meritst) through the following calculation:
Assumptions of the Instrumental Variables Framework
The credibility of instrumental variables results depends on satisfying three assumptions (Angrist & Pischke, 2009).
Relevance assumption
First, the relevance assumption states that the instrument affects the endogenous regressor. The instrumental variables coefficient estimate is based solely on variation in the endogenous regressor that is conditionally correlated with the instrument. Murray (2006) shows that instrumental variables estimates are biased in the direction of the bias of an ordinary least squares estimate and that the bias of the instrumental variables estimate is inversely related to the strength of the conditional correlation between the instrument and the endogenous regressor.
Several formal tests exist for the null hypothesis that the instrument is uncorrelated with the endogenous regressor (Murray, 2006). Table 2, column (1) presents the estimated results of the first-stage regression (Equation 2) of state merit-based aid in other states on nonresident enrollment. We found a negative relationship between state merit-based aid in other states and nonresident enrollment. Specifically, nonresident enrollment was found to decline by 0.5% as merit-based aid in other states increased by 1%. Both an F test of excluded instruments test (F = 13.13, p < .01) and a Kleibergen–Paap underidentification test (χ2 = 10.97, p < .01) indicated that the instrument identified relevant variation in nonresident enrollment. Furthermore, the Kleibergen–Paap weak identification test statistic was 13.13 (equivalent to the F test of excluded instruments in this just-identified single endogenous variable model) indicating that the maximum size distortion was likely between 10% (critical value of 16.38) and 15% (critical value of 8.96) (Stock & Yogo, 2005).
Relevance of the Instrument: First-Stage and Reduced Form Estimates
Note. Robust standard errors clustered at the institution level in parentheses. FTE = full-time equivalent.
p < .1. **p < .05. ***p < .01.
Independence assumption
Second, the independence assumption states that, conditional on covariates, the instrument is independent from omitted variables that affect resident enrollment (Angrist & Pischke, 2009). Satisfying the independence assumption requires that after including covariates, state merit-based financial aid generosity in other states is uncorrelated with omitted variables that affect resident enrollment. This would imply that state merit-based aid in other states is as good as randomly assigned in that it had no systematic relationship with omitted factors affecting resident enrollment.
The independence assumption cannot be tested directly (Wooldridge, 2002), rather, it rests on the plausibility of a logical argument. We argue that geographical boundaries that restrict the availability of state merit-based aid to residents implies that state merit-based aid in one state is as good as randomly assigned with respect to residents of other states.
Because states adopt merit-based aid to compete with other states for the “best and brightest” students (Doyle, 2006), one possible concern with our argument was the potential for an arms race in state merit-based aid expenditure. Under this scenario, increased state merit-based aid spending in Georgia may cause Alabama to increase state merit-based aid spending, which may affect resident enrollment at the University of Alabama. We mitigate this potential problem by controlling for state expenditure on merit- and need-based aid within the focal institution’s state. Second, we controlled for economic factors in an institution’s state because economic factors are correlated across state boundaries and may be related to generosity of merit-based aid in other states.
Exclusion restriction assumption
Third, the exclusion restriction assumption states that the only path through which the instrument affects the dependent variable is through the endogenous regressor (Angrist & Pischke, 2009). This assumption can be violated even when the random assignment assumption is fulfilled (Angrist & Pischke, 2009). Continuing with the previous example, the exclusion restriction assumption was satisfied if increased generosity in Georgia merit-based aid only affected resident enrollment at the University of Alabama through its effect on the number of Georgia students matriculating to the University of Alabama.
A potential pathway concern was raised by findings from Winters (2012). He found that when the population size of resident cohorts grew, institutions decreased nonresident enrollment and increased nonresident tuition. For this article, the pathway concern was that increases in state merit-based aid in other states caused resident enrollment in that state to increase to the point that universities became capacity constrained and restricted nonresident enrollment. For example, increased generosity of state merit-based aid in Georgia could have increased resident enrollment at Georgia institutions. In turn, Georgia institutions may have had less capacity to enroll Alabama students, which may have caused the number of Alabama residents who attended the University of Alabama to have increased.
The predicted bias associated with this pathway concern likely biases the coefficient of interest negatively, in the direction of finding a crowd-out effect. In the previously described pathway concern, when merit-based aid increased in Georgia, fewer students from Georgia were likely to attend the University of Alabama. Simultaneously, fewer students from Alabama went to Georgia higher education institutions due to increased capacity constraints at Georgia institutions, potentially increasing the likelihood of attendance at the University of Alabama. Thus, this potential alternative pathway predicts a potential negative relationship between nonresident and resident enrollment which was not due to the preferred instrumental variables pathway.
Although this alternative pathway was theoretically a concern for the instrumental variables identification strategy, we argue that the magnitude of this relationship was likely small. First, increased resident cohorts were only likely to crowd out nonresident enrollment at institutions operating near a physical or self-imposed capacity constraint. In fact, Winters (2012) has shown that larger cohorts of resident students were associated with a small decrease in nonresident enrollment at flagship public universities (a decrease of 24 nonresident students for an increase of 100 resident students) with no effect at nonflagship public universities. For this effect to be a concern for our analyses, the displaced students must have returned to their resident public research university as opposed to a different out-of-state public or any private institution. Using data from the Educational Longitudinal Study of 2002, students who were rejected by at least one out-of-state public university applied to an average of 5.3 higher education institutions, of which only 1.3 were a public institution in their state of residence (authors’ calculations). Furthermore, the same group of students was accepted to an average of 2.9 higher education institutions, of which 0.9 were in-state public universities. Combined, although we posit that the alternative pathway is a potential concern, the number of students who are displaced due to increased resident enrollment in a target state who then choose to enroll at their own state’s flagship institution was likely small.
Despite our argument that this pathway concern was small, we also attempted to mitigate the potential bias through the addition of two control variables. First, to control for the changing pressures on enrollment capacity at flagship public universities, we created two alternative measures of the enrollment capacity constraints flagship universities faced in other states. Based upon data available for the full sample utilized in this study, and similar to Winters (2012), we constructed a measure of flagship university enrollment capacity which was defined as the ratio of state population of 18-year-olds to resident freshman enrollment (this measure is weighted by distance and population in the methodology of the instrument). As this ratio increased, public research universities in other states faced increased capacity constraints due to increased demand from resident students, implying that they may have had less capacity to enroll nonresident students.
The values of the research university capacity proxy change for two primary reasons. First, the numerator represents the potential resident cohort of 18-year-olds, which as it grew, potentially placed pressure on the capacity of public universities to enroll nonresident students. The denominator, resident freshman enrollment at flagship universities captures changes in resident demand for flagship public universities. Combined, this construct captures changes to the demand for flagship public universities by the resident population. In particular, changes to the generosity of merit-based aid programs that decrease the relative cost of in-state public higher education were likely to cause potential capacity constraints, which could in turn have crowded out nonresident students.
Second, we control for nonresident tuition in other states (this measure is weighted by distance and population in the methodology of the instrument) to control for potential strategic behaviors of public institutions when faced with increased resident demand. As the demand for an institution from resident students increased, public institutions faced less pressure to meet enrollment quotas from nonresident enrollment and may have increased nonresident tuition in response (Groen & White, 2004; Winters, 2012).
Results
The Effect of Nonresident Enrollment on Resident Enrollment
Table 3 presents fixed effects (column 1) and instrumental variables (column 2) estimates of the relationship between nonresident enrollment and resident enrollment. Instrumental variables coefficients were estimated using a two-stage least squares estimator. In all models, robust standard errors were clustered at the institution level.
The Effect of Nonresident Enrollment on Resident Enrollment
Note. Robust standard errors clustered at the institution level in parentheses. FTE = full-time equivalent.
p < .1. **p < .05. ***p < .01.
The fixed effects estimates indicated that a 1% increase in nonresident enrollment was associated with a 0.1% increase in resident enrollment. The point estimate from the instrumental variables procedure was negative, although insignificantly different from zero. Differences in magnitude and statistical significance between the fixed effects and the instrumental variables results could be due to endogeneity bias in the fixed effects estimates or inefficiency in the instrumental variables estimates. Even if the positive point estimate from the fixed effects model were to be believed, the magnitude of this relationship was practically small and indicative of an enrollment management strategy in which nonresident enrollment was grown without restricting access for resident students.
Because a valid instrument only affects the dependent variable through its effect on the endogenous regressor, reduced form estimates of the effect of the instrument on the dependent variable are an important diagnostic check for instrumental variables analyses (Murray, 2006). If nonresident enrollment crowded out resident enrollment, we expect that increased state merit-based aid in other states to have had a positive effect on resident enrollment through decreased nonresident enrollment. Table 2, column 2 presents the reduced form estimates when the instrument was substituted for the endogenous regressor in the outcome equation (Equation 3). Consistent with the instrumental variables results, the reduced form results indicated that neighboring state merit-based aid did not have a direct relationship with resident enrollment.
Heterogeneity by Institutional Type
In the case of capacity constrained institutions (i.e., inelastic supply), we expected that any increase in nonresident enrollment must have come at the expense of resident enrollment. Unfortunately, capacity constraints and the relative importance of enrollment management goals are inherently unobservable within the context of the IPEDS data. Thus, we used prestige as a proxy for capacity constraints as prestigious universities were more likely to restrict enrollment capacity to maintain or increase prestige. We utilized the 2004 U.S. News and World Report National University Rankings as a proxy for prestige. The U.S. News methodology rewards institutions based upon measured enrollment attributes, such as spending per student, standardized test scores, and admissions rates. We separated institutions into two categories based on 2004 rankings: Prestigious institutions which were rated in the top 50 of national universities (17 of which are public universities), and less-prestigious institutions which included all other public research universities (the remaining 142 institutions in the sample).
Figure 2, which presents the equivalent of Figure 1 when the sample was restricted to the 17 prestigious public universities, visually supports the argument that nonresident students may have crowded out resident students at prestigious public universities. Specifically, beginning in 2006-2007, a decline in resident enrollment was observed while nonresident enrollment began to grow rapidly.
Average first-time freshman enrollment at prestigious (ranked in the top 50 of national universities by U.S. News and World) public research universities (N = 17).
To statistically investigate the heterogeneity of this response across institutional prestige, we extended the empirical framework to allow the coefficient on nonresident enrollment to vary by institutional prestige (represented by the indicator variable Dit). Specifically, we used an interaction model in which the only coefficient allowed to vary measured the relationship between nonresident and resident enrollment. The alternative approach would have been to estimate the model separately for prestigious and nonprestigious institutions. As the prestigious group is restricted to a sample size of 17 institutions, the interaction model was preferred to separate models as a way to save statistical power by restricting the relationship between control variables and the outcome to be the same for both prestigious and nonprestigious institutions. Although the prestigious group only contains 17 institutions, identification in a fixed effects estimator occurs within units and took advantage of the 18 years of data for most institutions within the sample (Wooldridge, 2002).
An instrumental variables model with an interacted endogenous variable needs two first-stage equations, and hence two instrumental variables. Specifically, we modeled the effect of the instrument (
Angrist and Krueger (1991) argued that a valid instrumental variable interacted with other plausibly exogenous factors becomes a set of valid instruments. In our case, we argue that the portion of the prestige of the institution, which does not vary across time (measured by U.S. News rankings in 2004), can be considered exogenous as this time-invariant prestige is controlled for through the institutional fixed effect and is uncorrelated with the idiosyncratic error term. Thus, conditional on the validity of the original instrument
A potential concern with the validity of the instrument within this framework is the geographic location of prestigious institutions of higher education relative to the location of state merit-based aid programs. State merit-based aid programs are overrepresented in the southern portion of the United States; thus, institutions closer to the south are likely to realize larger variation in the instrument. Of the 17 prestigious institutions, over a third are located in the southern portion of the United States (College of William and Mary, Georgia Institute of Technology, University of Florida, University of North Carolina–Chapel Hill, University of Texas–Austin, and University of Virginia). More importantly, Table 1 demonstrates the similarity of the means of the instrument for the prestigious institutions (M = 73.5) and the less-prestigious institutions (M = 74.0). Furthermore, the within-institution standard deviations of the instrument were similar, and larger than the between-institution standard deviations, for the prestigious institutions (within SD = 28.8, between SD = 17.4) and the less-prestigious institutions (within SD = 29.8, between SD = 23.5).
Table 4 presents the first-stage results for Equation 6 (column 1) and Equation 7 (column 2). In general, the first-stage estimates were consistent with both directional and statistical significance expectations. At less-prestigious institutions, a 1% increase in state merit-based financial aid in other states led to a 0.55% decrease in nonresident enrollment. For the prestigious institutions, a 1% increase in state merit-based financial aid in other states led to a 0.22% increase in nonresident enrollment (calculated through the addition of the four coefficients associated with state merit-based aid, 0.215 = −0.550 + 0.304 + −0.142 + 0.603), although the coefficients were jointly insignificantly different from zero (p > .1). It is not surprising that neighboring state merit-based financial aid has an insignificant effect on nonresident enrollment at prestigious institutions because demand by Georgia residents for prestigious out-of-state institutions (e.g., University of Virginia) is less likely to be sensitive to changes in Georgia merit-based aid generosity than demand by Georgia residents for less-prestigious out-of-state institutions. The Kleibergen–Paap weak identification test statistic was 5.43, indicating that the maximum relative bias was likely between 10% (critical value of 7.03) and 15% (critical value of 4.58) (Stock & Yogo, 2005).
Interaction Model: Relevance of the Instrument: First-Stage and Reduced Form Estimates
Note. Robust standard errors clustered at the institution level in parentheses.
p < .1. **p < .05. ***p < .01.
The outcome equation (Equation 8) for the interacted instrumental variables equation contained the predicted values from both first-stage equations.
Table 5 presents fixed effects (column 1) and instrumental variables (column 2) estimates of the relationship between nonresident enrollment and resident enrollment when allowing the relationship to differ between more- and less-prestigious institutions. For the less-prestigious institutions, a 1% increase in nonresident enrollment was associated with a 0.1% increase in resident enrollment in the fixed effects estimator and a positive but insignificant relationship in the instrumental variables estimator.
Interaction Model: The Effect of Nonresident Enrollment on Resident Enrollment
Note. Robust standard errors clustered at the institution level in parentheses.
p < .1. **p < .05. ***p < .01.
The coefficient on the interaction between nonresident enrollment and the prestige indicator indicated whether the effect of nonresident enrollment was different for prestigious institutions and less-prestigious institutions. For the fixed effects model (column 1), this interaction coefficient was negative and insignificant, and the combined marginal effect for prestigious institutions indicated a positive but insignificant point estimate (0.07, p = .39). For the instrumental variables model (column 2), the interaction effect was negative and statistically significant. The marginal effect for prestigious institutions indicated that 1% increase in nonresident enrollment led to a 0.18% (=0.016 − 0.197, p = .07) decrease in resident enrollment.
The instrumental variables estimate for prestigious institutions suggested a crowding-out effect on resident enrollment as institutions increased nonresident enrollment. For the average prestigious public research institution during the final year of our sample (2014–2015), the estimated model would predict that an increase of 15 nonresident freshmen (1% of 1,467) would decrease resident freshman enrollment by 9 (0.2% of 4,277). Alternatively, between 2012–2013 and 2014–2015, nonresident enrollment grew by an average of 5.4% per year, which translates to a predicted increase of 80 nonresident students likely having crowded out 46 resident students. Thus, although the coefficients which are expressed as elasticities appear to be relatively small, the estimated crowd-out effect is practically important.
Robustness to Alternative Specifications
Alternative Instruments
An important concern with the instrumental variables estimation strategy were potential biases in the coefficient of interest due to a weak instrument. In the models estimated previously, weak identification tests using Stock and Yogo (2005) critical values indicated that the maximum relative size distortion associated with our models was less than 15%. The failure to reject a hypothesis of less than 10% may lead one to question the strength of our instrumental variable. To assess the potential biases in our estimates, we re-estimated the primary models with alternative constructions of the instrumental variable.
Table 6 presents the coefficient of interest, the effect of nonresident enrollment on resident enrollment, for alternative constructions of the instrumental variable. Column 1 represents the equivalent model as presented in Table 3, with columns 2 through 7 presenting the re-estimation of this model with alternative sets of instruments. Panel B represents the same progression of models for the interaction model presented in Table 5.
Robustness of Findings to Alternative Choices of Instruments
Note. Robust standard errors clustered at the institution level in parentheses.
The percentage represents the maximum relative size distortion based upon critical values in Stock and Yogo (2005).
p < .1. **p < .05. ***p < .01.
Estimates were robust to alternative distance weighting schemes in the construction of the merit-based aid instrument. Specifically, we allowed the weighting scheme to be more sensitive to distance by squaring distance (column 2) and less sensitive to distance by taking the square root of distance (column 3). For both alternative instrument distance treatments, estimates of the relationship between nonresident and resident enrollment were qualitatively similar to the linear treatment. However, in both cases, the underidentification and weak identification statistics indicated that the alternatively weighted instruments did not predict nonresident enrollment as well as the linearly weighted instrument.
Following the approach of Angrist and Krueger (1991), we interacted the instrument with plausibly exogenous factors in an effort to increase the relative strength of the instrument. Columns 4 and 5 present results when the merit-based aid instrument was interacted with indicators of the institution’s census division (eight divisions) and 2004 U.S. News and World Report tier (four tiers). In general, the point estimates of the coefficient of interest were qualitatively similar, although in the census division model, the interaction term is not statistically significant. In each case, although the overall ability of the set of instruments to explain the endogenous regressor improved as measured by the underidentification test, the maximum relative size distortion increased due to the added instrumental variables.
Following the same geography-based logic made for the choice of state merit-based aid as a valid instrument, one could argue for the inclusion of state needs-based aid as an alternative instrument. Estimates using both state merit- and need-based aid as instruments (column 6) found qualitatively similar results as models with solely merit-based aid as an instrument. The inclusion of needs-based aid decreased the strength of the instruments in explaining nonresident enrollment and increased the maximum relative size distortion of the instrumental variables estimate.
Following the approach of Winters (2012), we included the state population of 18-year-olds in other states in addition to state merit-based aid in the instrument set (column 7). Estimates of the relationship between nonresident and resident enrollment were similar to our base model. Similar to the addition of needs-based aid, measures of the strength of the relevance of the instrument indicated the combined population and merit-based aid instrument set was weaker than using merit-based aid alone.
Alternative Pathway Controls
Another concern with the instrumental variables estimates was the potential presence of alternative pathways between the instrument and outcome. Drawing from the findings of Winters (2012), earlier we described the following alternative pathway: increased merit based-aid in Georgia positively affected demand at in-state institutions by Georgia residents, which crowded out Alabama residents from Georgia institutions and, in turn, increased the number of Alabama residents attending in-state institutions. Our primary analyses attempted to mitigate this pathway concern by controlling for a proxy for research university capacity and nonresident tuition price.
Starting in 2002, IPEDS began to provide more detailed information regarding institutional selectivity measures. As a robustness check, we include the weighted average (based upon distance from the focal institution to competitor public research universities in other states) acceptance rate in place of the state-level measures of public flagship university capacity. We argue that acceptance rate is a better proxy for enrollment capacity constraints, as opposed to the ratio of 18-year-olds to resident enrollment, as it more directly measured the ratio of student admitted compared with those who would seek admission. Admissions rates directly measure whether an institution is restricting enrollment relative to the number of applications received. If an increase in state merit-based aid leads to increased demand for a particular institution (i.e., increased applications) and that institution did not increase their enrollment, then admissions rates would have declined. By controlling for changes in admissions rates, the effect of increased merit-based aid on the demand for resident students which potentially crowds out nonresident students can be mitigated. The tradeoff for including better measures of institutional selectivity was the reduced sample size due to data availability (about 30% of our sample was lost).
Table 7 presents the results of this sensitivity check where column 1 presents the findings from the preferred specification, column 2 presents the preferred specification restricted to the years for which the new pathways controls were available, and column 3 presents the results using the new pathways controls in place of the original controls. For the pooled model, results were similar across all three specifications. The primary difference was that the instruments were considerably weaker when the sample was restricted to the shorter time period for both specifications of control variables. For the interaction model, the negative relationship was found for prestigious institutions in all three specifications. The point estimates were more negative in both of the shorter period models, although the corresponding confidence intervals were much larger.
Robustness of Findings to Alternative Pathway Controls
Note. Robust standard errors clustered at the institution level in parentheses.
The original pathway controls include weighted (by population and distance) average state-level measures for nonresident tuition and fees and research university capacity (18-year-old population/resident enrollment at public research universities).
The revised pathway controls include weighted (by distance) average institution-level measures for the percentage of applicants admitted, in-state tuition and fees, nonresident tuition and fees, and institutional grants.
The percentage represents the maximum relative size distortion based upon critical values in Stock and Yogo (2005).
p < .1. **p < .05. ***p < .01.
Alternative Samples
Table 8 presents results comparing the coefficient of interest for models that utilized observations from years where the IPEDS Residence and Migration survey were voluntary (columns 1 and 2) compared with specifications where those observations were dropped (columns 3 and 4). The findings were qualitatively similar across both samples. The key difference of interest resides in the instrumental variables estimates of the interaction model, where the prestigious interaction coefficient was not statistically significantly different from zero, although the overall point estimate was similar. Thus, while statistical significance was lost, the overall estimate of the effect size remained relatively stable across samples.
Robustness of Alternative Samples
Note. Robust standard errors clustered at the institution level in parentheses.
The percentage represents the maximum relative size distortion based upon critical values in Stock and Yogo (2005).
p < .1. **p < .05. ***p < .01.
One concern in the models that allows the relationship to differ by institutional prestige is that the prestigious group was overrepresented by institutions in California. Specifically, although California institutions represented only nine of the 159 total institutions, they represented 6 of the 17 prestigious institutions. Table 8 presents results where California institutions were dropped from the analytical sample (columns 5 and 6). The results were qualitatively similar when California institutions were dropped, with a stronger negative relationship estimated between nonresident and resident enrollment for the prestigious group of institutions.
Discussion
During the 2000s, resident enrollment growth at public research universities started to stagnate while nonresident enrollment growth increased. More recently, state policymakers became concerned that the increased emphasis on the enrollment of nonresident students could be coming at the expense of access for resident students (Burd, 2015a, 2015b). This study provided empirical evidence as to whether increased enrollment growth of nonresident students at public research universities crowded out access for resident students.
For the full sample of research universities, instrumental variables results did not indicate that enrollment growth of nonresident students crowded out the enrollment of resident students. Thus, the evidence suggests that public research universities likely pursued an enrollment strategy to increase nonresident students independent of their resident enrollment strategy. However, at prestigious public research universities, evidence was found that increased nonresident enrollment caused a decrease in resident enrollment. At the average prestigious university between 2012–2013 and 2014–2015, the model predicted that 46 resident students were crowded out by the average annual increase of 80 nonresident students. These findings were consistent with prestigious universities having inelastic short-term supply, in which an increase in nonresident enrollment must come at the expense of a resident student.
We urge caution when interpreting the findings due to several limitations. First, we remain somewhat concerned about the overall strength of the merit-based aid instrument to predict nonresident enrollment. In the preferred specifications, weak identification tests indicated that the maximum size distortion of our instrumental variables estimates was less than 15% when ideally a strong instrument would have been less than 10%. Thus, some potential bias is likely to remain in our instrumental variables estimates, which is particularly important when interpreting the findings for prestigious institutions. A second limitation exists due to the regional nature of large-scale state merit-based aid programs. Although state merit-based aid programs exist across the country, the largest are concentrated in the southern part of the United States. Thus, variation in the instrumental variable was largest in the south which may limit the generalizability of the findings as instrumental variables estimate a local average treatment effect.
The findings suggest that policymakers should not be concerned that nonresident enrollment growth at public research universities is crowding out resident enrollment, except perhaps at the most prestigious universities. Thus, policy changes aimed at the growth of nonresident enrollment at public universities, such as nonresident enrollment caps, are unlikely to increase access for resident enrollment. Furthermore, restricting nonresident student enrollment may be against state fiscal interests as nonresident students have been found to pay more in net tuition and future state taxes than resident students (Groen & White, 2004). This is particularly true for the prestigious institutions, which are most likely to be attracting high academic ability students to their state. Given the significant increase in interstate migration for higher education, more research is needed to understand whether such policies are beneficial to, or may harm, state economic development goals.
Given the recent interest in both implementing nonresident enrollment caps (e.g., California) and removing nonresident enrollment caps (i.e., Wisconsin), more research is needed to understand the distributional consequences of such policies for state higher education systems. Our finding that nonresident enrollment crowded out resident enrollment at prestigious public universities is consistent with growing concern that access for resident and low-income students to flagship public universities is being compromised as institutions increasingly pursue admissions policies aimed at tuition revenue and/or academic profile. Previous research has shown that the diversity of the student body (e.g., low-income and underrepresented minority students) declined as prestigious public universities increased the share of nonresident students on their campuses (Jaquette et al., 2016). However, research has not shown how educational attainment changes for students who are crowded out of their state flagship public universities. Future research should seek to understand the shifting enrollment patterns of state residents as their public universities prioritize nonresident enrollment.
Finally, future research should investigate the benefits and consequences of the pursuit of nonresident enrollment at public research universities. For example, research has shown that spending on instruction and student services is positively associated with graduation rates (Webber, 2012; Webber & Ehrenberg, 2010). Public universities often argue that nonresident enrollment growth is necessary to maintain quality instruction and student services amid state funding cuts. However, prior research has not addressed how public universities are spending revenue from nonresident tuition. Therefore, more research is needed to investigate whether universities are expending these revenues on practices shown to improve educational outcomes as opposed to expenditures on nonacademic improvement, such as recreation centers, athletics, and other facilities.
Footnotes
Acknowledgements
We thank two anonymous reviewers, Steve DesJardins, Steve Porter, and seminar participants at the 2016 Association for Education Finance and Policy (AEFP) Conference, the University of Georgia, and the University of Missouri for valuable feedback on earlier drafts.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Authors
BRADLEY R. CURS is an associate professor of higher education at the University of Missouri. His research uses tools from the discipline of economics to understand the effects of educational policy on students and institutions. Recently, he has focused on issues of education access, equity, and success across three domains: the efficacy of financial aid programs, the behaviors of educational institutions, and precollege readiness behaviors. Address: 202 Hill Hall, University of Missouri, Columbia, MO 65211. Email:
OZAN JAQUETTE is an assistant professor of education at the University of California, Los Angeles. He studies the organizational behavior of colleges and universities. His research program analyzes how colleges and universities change behavior to generate enrollment from desired student populations, with a focus on tensions between stated commitments to access and enrollment behaviors motivated by the pursuits of revenue and academic prestige. Address: Moore Hall 3038, 405 Hilgard Avenue, Los Angeles, CA 90095. Email: