Abstract

In the wake of declining state support for higher education, many state leaders have adopted lottery earmark policies, which designate lottery revenue to higher education budgets as an alternative funding mechanism. However, despite the ubiquity of lottery earmarks for higher education, it remains unclear whether this new source of revenue serves to supplement or supplant state funding for higher education. In this paper, we use a difference-in-differences design for the years 1990–2009 to estimate the impact on state appropriations and state financial aid levels of designating lottery earmark funding to higher education. Main findings indicate that lottery earmark policies are associated with a 5 percent increase in higher education appropriations, and a 135 percent increase in merit-based financial aid. However, lottery earmarks are also associated with a decrease in need-based financial aid of approximately 12 percent. These findings have serious distributional implications that should be considered when state lawmakers adopt lottery earmark policies for higher education.

1.  Introduction

Over the last few decades, the number of states allocating lottery earmark funds to support higher education has expanded significantly. By 2009, twenty-five states dedicated some portion of state lottery revenues to support higher education appropriations, state financial aid, or capital improvement projects for public universities. This trend in state higher education budgets raises important questions for scholars in higher education finance. For instance, although lottery earmarks are marketed to the public as providing supplementary nontax base funding, scholars studying similar budget areas, such as K–12 education, have questioned whether the utilization of lottery funds supplants state funding for education (Dye and McGuire 1992; Miller and Pierce 1997; Covert 2014). However, previous studies have yet to address the impact of lottery earmarks in the unique and consequential context of higher education funding.

To clarify the impact of lottery earmarks on state funding for higher education, we test whether the adoption of lottery earmarks supplements or supplants state higher education funding levels. Specifically, using annual state-level data from 1990–2009, we leverage a difference-in-differences (DID) approach to estimate how adopting lottery earmark policies influences state higher education appropriations, state merit-based financial aid, and state need-based financial aid.1 In addition, we use a variety of empirical approaches and robustness checks to improve confidence in these estimates, including a dynamic panel model, placebo tests, and a flexible DID model that allows the estimated impacts to vary across years relative to the year of implementation.

The empirical analysis reveals that the adoption of lottery earmarks for higher education supplements both state appropriations and merit-based financial aid levels. Specifically, our estimates reveal that the adoption of lottery earmarks is associated with an increase in higher education appropriations by approximately 5 percent, and merit-based financial aid by approximately 135 percent, on average. However, this increase is concentrated in states that have used lottery earmarks to create scholarship programs. These states almost doubled appropriations to merit-based aid, whereas states without a specific lottery scholarship decreased funding by approximately 33 percent. Lottery earmarks are also associated with a decrease in need-based financial aid of approximately 12 percent, potentially highlighting a supplanting impact on financial aid targeting the economically disadvantaged.

This paper proceeds with a description of the higher education financial landscape, along with a discussion of the processes by which lottery revenues are dedicated to state higher education budgets. Next, to inform our analysis, we leverage insights from previous literature on public finance and on the estimated impact of lottery earmarks on K–12 education budgets. Then, we summarize the unique panel dataset underlying this analysis and detail the empirical approach, as well as potential threats to validity of the research design. Lastly, we present the findings of the empirical analysis, conduct robustness checks, and discuss implications for policy makers and scholars.

2.  Higher Education Funding Mechanisms

Research in higher education finance has established the overall decline in state support for public universities over time. This decline is due in part to public resistance to increasing taxes, the unique ability of universities to generate their own revenue through tuition increases, and the discretionary nature of higher education budgets that moves it to the end of the state budgeting queue (Cheslock and Gianneschi 2008). As a result, state appropriations to higher education are particularly susceptible to cuts in economic recessions, and once the budgets have been cut, it is unlikely that state leaders will reinvest in appropriations at pre-recession levels (Zumeta 2012). These conditions create unique financial challenges for public universities facing disproportionate cuts to state higher education appropriations, which often contribute to substantial tuition hikes to make up for funding shortfalls (Martin and Gillen 2011; McLendon, Tandberg, and Hillman 2014; Perna and Finney 2014, p. 19).

Indeed, despite the cyclical fluctuations over time, public universities receive less funding from state appropriations and rely more heavily on tuition revenue, which has serious implications for college affordability.2 This is not a recently emerging trend; in fact, from the 1970s to the mid 2000s, state appropriations fell by 30 percent, in spite of the increasing cost of administering public higher education (Archibald and Feldman 2006). Moreover, the policy implications of decreasing state appropriations amidst increasing student demand and cost have far-reaching economic consequences (McLendon, Hearn, and Mokher 2009). By increasing tuition in response to declining state appropriations, public universities shift financial burden onto students and families. This shifting financial burden can undermine institutional missions by limiting access, retention, and graduation for economically disadvantaged students, as well as threatening the economic well-being of many industries that rely upon educated workers (Weisbrod, Ballou, and Asch 2010; Perna and Finney 2014).

As a result of these problematic shifts in higher education finance, state leaders have pursued alternative funding mechanisms aimed at providing supplementary funding to higher education. Among the most ubiquitous of these policies is the implementation of lottery earmarks for higher education appropriations, grant and scholarship programs, or capital improvement projects. As of 2009, twenty-five of the forty-five states that operate lotteries adopted policies designating lottery earmark funds to higher education. Despite the widespread nature of state funding policies designating lottery earmarks to higher education, previous studies have focused on the implications of lottery earmarks for K–12 education and have largely ignored their impacts on higher education. This study remedies this oversight by thoroughly investigating the relationship between the adoption of lottery earmarks for higher education and state funding through appropriations and state need-based and merit-based financial aid. In this investigation, we begin with an in-depth examination of the variation in lottery earmarks for higher education across states.

Variation in Lottery Earmarks

To effectively characterize the variation across the twenty-five states with lottery earmarks for higher education, we contacted officials in each state legislature and higher education governing institution regarding their state policies governing the allocation of lottery earmarks for higher education. From these conversations, we have identified three key types of state policies that allocate lottery revenue to higher education.3 First, there are states that earmark lottery revenue directly to specialized funds that support higher education. For example, Idaho dedicates a portion of lottery revenues to its Permanent Building Fund, which in turn funds the maintenance and establishment of university campus buildings, among other higher education capital projects. Second, there are states such as Maine that transfer lottery revenue from the general fund elsewhere, ultimately designating that revenue to support higher education budgets.

In the third key policy type, states allocate lottery earmark funds to scholarship programs. For instance, in 1993, Georgia adopted the HOPE scholarship program, which is perhaps the most well-known and well-researched program that dedicates lottery revenue to student financial aid (Dynarski 2002; Long 2004; Cornwell, Mustard, and Sridhar 2006; Brady and Pijanowski 2007). These programs have become more abundant in recent years, with seventeen states out of the twenty-five with lottery earmarks for higher education dedicating some portion of the revenue to state financial aid programs during our time period of study. The variation in policy structure is described in table 1 for each of the states that have adopted lottery earmarks for higher education in the first year of adoption.4

Table 1.
Overview of Lottery Earmark Support of Higher Education
StateYear of State Lottery Policy ImplementationYears of AmendmentsDescription of Policy ImpactsDescription of Scholarship Programs and Type(s)a
New Jersey 1970 NA Funds a special fund for K—12 and higher education NA
Maryland 1973 NA Funds dedicated to Maryland Education Trust Fund for community colleges and four-year public higher education. NA
Maine 1974 1996 Funds transferred to the General Fund which is then used to fund K—12, higher education, among other programs. Since 1996 transfers funds to Outdoor Heritage Fund NA
California 1984 2000; 2010 Funds used for acquisition of property, construction of facilities, finance research, other noninstructional purposes. Loan forgiveness programs, need-based aid for summer remediation programs.
Montana 1987 NA Funds transferred to general fund; excess over previous fiscal year is transferred to Montana STEM Scholarship Program Montana STEM Scholarship Program (Merit)
Arizona 1980 1990; 1993; 1996; 2010; 2015 Funds state parks, health and social services programs, education (University bonds and Native American Dual Enrollment programs), local transit and counties NA
Idaho 1990 NA Funds used for construction of buildings including at higher education institutions NA
Georgia 1993 NA Funds tuition grants, scholarships, loans; Georgia Prekindergarten program; capital outlay projects for K—12 and higher education HOPE Scholarship (Merit-based)
Nebraska 1993 2003 Funds Nebraska State Fair; education improvement fund; Opportunity Grant Fund (since 2003); environmental trust; compulsive gamblers assistance fund Opportunity Grant Fund (need-based scholarships)
Delaware 1995 NA Appropriated to General Fund and then dedicated to higher education NA
Oregon 1995 NA Funds academic scholarships and intercollegiate athletics Oregon Lottery Graduate Scholarship (Need and merit based); others distributed through individual universities
New Mexico (NM) 1996 2014 Funds a tuition-based scholarship that is available to all first-time, full-time college students who graduated from a NM high school. Legislative Lottery Scholarship
Florida 1997 NA Funds the Educational Enhancement Trust Fund Bright Futures Scholarship (Merit-based)
Missouri 1997 NA Funds used to fund budgets at various state universities, colleges, and community colleges A+ Scholarship (Merit-based); Access Missouri (Need-based); Minority Teaching Scholarship; Marguerite Ross Barnett Scholarship Program
South Dakota 1997 NA Transferred to General Fund, nearly half of which supports K—12 and higher education NA
South Carolina 1998 NA Funds a variety of scholarship programs, technology improvements, endowed chairs, and other programs Palmetto Fellows (Merit-based); Life Scholarship (Merit-based); HOPE scholarship (Merit-based); State Need-based Grant
Kentucky 1999 NA Funds various scholarship programs as well as K-12 literacy initiatives Kentucky Excellence Award (Need and merit-based); College Access Program (need-based); Kentucky Tuition Grant (need-based)
West Virginia (WV) 2000 2002 Funds university base budgets except for WVU and Marshall (line item) as well as PROMISE Scholarship PROMISE Scholarship (Merit-based)
Washington 2000 2004; 2010 Funds used for education construction and maintenance as well as Opportunity Pathways Account and merit scholarships Opportunity Pathways Fund (Need-based); two unnamed Merit-based scholarship programs
Tennessee 2004 NA Funds various scholarship programs HOPE Scholarship (Merit-based); General Assembly Merit scholarship; Aspire Award (Merit and Need based); Hope Access Grant (Need and Merit based)
Oklahoma 2005 NA Funds tuition grants, loans and scholarships, endowed chairs, and facility/technology expansions and improvements Oklahoma Academic Scholars Program (Merit-based)
North Carolina 2006 NA Funds scholarships through the State Educational Assistance Authority North Carolina Education Lottery Scholarship (Need-based)
Connecticut 2007 NA Appropriated to General Fund and then dedicated to supplement higher education budgets NA
Iowa 2008 NA Funds are appropriated through General Fund, Veterans Assistance Fund, and Vision Iowa Tuition Assistance offered through Veterans Assistance Fund
Arkansas 2009 NA Funds a special fund that is separate from the General Assembly for lottery administration and scholarships Arkansas Academic Challenge (Merit-based)
StateYear of State Lottery Policy ImplementationYears of AmendmentsDescription of Policy ImpactsDescription of Scholarship Programs and Type(s)a
New Jersey 1970 NA Funds a special fund for K—12 and higher education NA
Maryland 1973 NA Funds dedicated to Maryland Education Trust Fund for community colleges and four-year public higher education. NA
Maine 1974 1996 Funds transferred to the General Fund which is then used to fund K—12, higher education, among other programs. Since 1996 transfers funds to Outdoor Heritage Fund NA
California 1984 2000; 2010 Funds used for acquisition of property, construction of facilities, finance research, other noninstructional purposes. Loan forgiveness programs, need-based aid for summer remediation programs.
Montana 1987 NA Funds transferred to general fund; excess over previous fiscal year is transferred to Montana STEM Scholarship Program Montana STEM Scholarship Program (Merit)
Arizona 1980 1990; 1993; 1996; 2010; 2015 Funds state parks, health and social services programs, education (University bonds and Native American Dual Enrollment programs), local transit and counties NA
Idaho 1990 NA Funds used for construction of buildings including at higher education institutions NA
Georgia 1993 NA Funds tuition grants, scholarships, loans; Georgia Prekindergarten program; capital outlay projects for K—12 and higher education HOPE Scholarship (Merit-based)
Nebraska 1993 2003 Funds Nebraska State Fair; education improvement fund; Opportunity Grant Fund (since 2003); environmental trust; compulsive gamblers assistance fund Opportunity Grant Fund (need-based scholarships)
Delaware 1995 NA Appropriated to General Fund and then dedicated to higher education NA
Oregon 1995 NA Funds academic scholarships and intercollegiate athletics Oregon Lottery Graduate Scholarship (Need and merit based); others distributed through individual universities
New Mexico (NM) 1996 2014 Funds a tuition-based scholarship that is available to all first-time, full-time college students who graduated from a NM high school. Legislative Lottery Scholarship
Florida 1997 NA Funds the Educational Enhancement Trust Fund Bright Futures Scholarship (Merit-based)
Missouri 1997 NA Funds used to fund budgets at various state universities, colleges, and community colleges A+ Scholarship (Merit-based); Access Missouri (Need-based); Minority Teaching Scholarship; Marguerite Ross Barnett Scholarship Program
South Dakota 1997 NA Transferred to General Fund, nearly half of which supports K—12 and higher education NA
South Carolina 1998 NA Funds a variety of scholarship programs, technology improvements, endowed chairs, and other programs Palmetto Fellows (Merit-based); Life Scholarship (Merit-based); HOPE scholarship (Merit-based); State Need-based Grant
Kentucky 1999 NA Funds various scholarship programs as well as K-12 literacy initiatives Kentucky Excellence Award (Need and merit-based); College Access Program (need-based); Kentucky Tuition Grant (need-based)
West Virginia (WV) 2000 2002 Funds university base budgets except for WVU and Marshall (line item) as well as PROMISE Scholarship PROMISE Scholarship (Merit-based)
Washington 2000 2004; 2010 Funds used for education construction and maintenance as well as Opportunity Pathways Account and merit scholarships Opportunity Pathways Fund (Need-based); two unnamed Merit-based scholarship programs
Tennessee 2004 NA Funds various scholarship programs HOPE Scholarship (Merit-based); General Assembly Merit scholarship; Aspire Award (Merit and Need based); Hope Access Grant (Need and Merit based)
Oklahoma 2005 NA Funds tuition grants, loans and scholarships, endowed chairs, and facility/technology expansions and improvements Oklahoma Academic Scholars Program (Merit-based)
North Carolina 2006 NA Funds scholarships through the State Educational Assistance Authority North Carolina Education Lottery Scholarship (Need-based)
Connecticut 2007 NA Appropriated to General Fund and then dedicated to supplement higher education budgets NA
Iowa 2008 NA Funds are appropriated through General Fund, Veterans Assistance Fund, and Vision Iowa Tuition Assistance offered through Veterans Assistance Fund
Arkansas 2009 NA Funds a special fund that is separate from the General Assembly for lottery administration and scholarships Arkansas Academic Challenge (Merit-based)

Notes: NA = not applicable.

aScholarships and programs listed were present at year of implementation. Implementation years and funding dedications were collected from state agency Web sites and correspondence with higher education governing boards and state legislative research agencies.

Furthermore, utilizing state lottery earmarks to defray the costs of state student aid programs has serious implications for college affordability for many disadvantaged students (Dynarski 2000; Hanushek, Leung, and Yilmaz 2014). Indeed, in spite of the large role the federal government plays in the financial aid process through Pell Grants and federal student loans, state scholarships and grant programs have also played an essential role in improving access and affordability for vulnerable and low-income students (Toutkoushian and Shafiq 2009, p. 44).5 However, although some lottery-funded scholarships are structured to help the students who need help the most, many states have recently shifted away from need-based financial aid in pursuit of expanded merit-based financial aid (Toutkoushian and Shafiq 2009; McLendon, Tandberg, and Hillman 2014). This shift has serious implications for affordability. In fact, whereas most merit-based scholarship programs have been shown to widen the gap in college attendance between advantaged and disadvantaged students (Dynarski 2000), need-based aid policies are shown to be a much more efficient option at boosting not only the educational opportunities of high-ability low- to moderate-income children, but also the efficiency of the overall economy (Hanushek, Leung, and Yilmaz 2014). Furthermore, research has revealed that need-based and merit-based financial aid are zero-sum, which makes it especially important to understand the balance between state investment of lottery revenue in these two forms of financial aid (McLendon, Tandberg, and Hillman 2014). Taken together, given the wide-ranging implications of state appropriations, as well as merit- and need-based financial aid for college affordability, it is essential to better understand the impact of designating lottery earmarks to these foundational higher education funding sources.

3.  Conceptualizing the Impacts of State Lotteries on Higher Education Funding

In order to build a firm conceptual understanding on the likely relationships between state lottery earmarks and higher education funding, we draw on two main sources of previous literature to inform our analysis: first, we leverage public finance literature on the “flypaper effect” and second, we provide a detailed investigation into the literature estimating the impact of lottery funds on education spending. These two sources of literature form the theoretical foundation for our unique exploration of lottery earmarks and higher education appropriations and state grant aid.

An array of public finance scholars has examined the “flypaper effect,” or the extent to which federal government grants to state and local public agencies augment these institutions’ levels of overall public spending. More specifically, the flypaper effect posits that government income in the form of grants leads to higher public agency spending than equivalent institutional revenue. However, these studies also highlight some fungibility in grants (Hines and Thaler 1995; Gordon 2004). Education finance literature reveals that the money sticks where it hits, but state and local revenue that is diverted to educational institutions does not necessarily beget higher levels of education spending. This is due to manipulation by elected officials in entrenched political institutions (Barrow and Rouse 2004; Inman 2008). For example, Dye and McGuire (1992) have determined that across four different state spending categories (including K–12 education), a one-dollar increase in lottery earmarks is associated with less than a one-dollar increase in general revenue spending. These disparate trends demonstrate that novel forms of revenue, such as lottery revenues, do not necessarily precipitate higher demand for education spending; rather, lottery policy implementation creates a new revenue stream to supplement federal grants and other sources of income. These dynamics are also evident in a recent examination of changes to K–12 education spending following implementation of tax increment financing (TIF) districts, in which Nguyen-Hoang (2014) found that new supplementary property taxes did not have anticipated spillover effects that would have increased counties’ and school districts’ educational expenditures. In Iowa's case, from 2001 to 2011, greater use of TIF was associated with lower educational expenditures after TIF districts expired than before TIF policy implementation. Similar to state legislatures’ claims that adopting lottery earmarks will elevate long-term education spending at both the state and institutional levels, policy makers justify TIFs as a mechanism for encouraging municipalities’ long-term public investment in specific locales. However, Nguyen-Hoang's findings suggest that TIF revenues are only temporarily supplementary for counties and school districts. In sum, this literature suggests that lottery earmarks are unlikely to translate to a dollar-for-dollar increase in higher education spending, since political institutions may manipulate earmarks.

These dynamics in revenue source variability and public spending are also evident in the literature investigating the impact of lottery earmarks on K–12 education spending. More specifically, education finance scholars have analyzed trends in per-student educational expenditures before and after the implementation of lottery earmarks for K–12 education to examine whether the same fungibility is evident in earmarking. The evidence on this question remains mixed; some scholars find that K–12 educational spending was not supplemented by earmarking (Mikesell and Zorn 1986; Borg and Mason 1990; Spindler 1995; Vance and Alsikafi 1999), whereas other studies find that lottery profits do boost state aid to local school districts but that these funds are fungible (Dee 2004; Evans and Zhang 2007). Finally, another group of scholars argues that instead of serving as a supplementary fund, lottery earmark revenue is utilized by state officials to decrease general appropriations to education over time (Miller and Pierce 1997; Covert 2014). Specifically, Miller and Pierce (1997) find that from 1966–1990, there is an initial increase in K–12 education expenditures as a result of lottery earmarks but the rate of growth in education spending declines over time (p. 40). However, in two recent studies with arguably the most compelling designs, Novarro (2005) and Evans and Zhang (2007) utilize DID models along with two-stage least squares regression models to uncover precise estimates of the fungibility in lottery earmark funds for K–12 education funding. Novarro finds that a dollar of lottery profits earmarked for K–12 education increased spending per pupil by 79 cents. Evans and Zhang find that out of a dollar of earmarked lottery revenue, 50 to 70 cents are appropriated to local school districts, and the remaining 30 to 50 cents in the general fund are diverted to other purposes. Our study builds on this literature by utilizing a DID and instrumental variables technique to address whether these findings hold in the unique and previously unexplored context of higher education finance. Taken together, the literature on “flypaper effects” and earmarking lottery funds for K–12 education suggests that lottery earmarks may translate to increases in higher education spending but are likely to be fungible.

Lottery Earmarks in the Context of Higher Education

Whereas the fungibility of state lottery earmarks has been studied extensively in other state budget areas (Erekson et al. 2002; Novarro 2005; Evans and Zhang 2007), higher education budgets have not previously been examined. This is an important oversight because higher education budgets are especially susceptible to variation due to both economic and political factors. In fact, appropriations to higher education have been described as the most elastic category of state budgets (White and Musser 1978). Furthermore, in many states there are few, if any, safeguards in place to ensure that state leaders use state lotteries to supplement available revenue sources and not substitute, or supplant, the resources at their disposal. For instance, while Title I requires school districts and other local education agency recipients of federal grants and other monies to match or otherwise supplement funds available from other sources by disallowing agency reallocation of Title I funds for specific activities (Arnone 2003), a similar provision does not exist for state lotteries and higher education (Shaul 2002). With that in mind, the impact of allocating lottery earmark funds to state higher education budgets requires closer examination.

Finally, despite the lack of evidence on the impact of lottery earmarks on higher education spending to date, a recent article sheds light on the impact of earmarking, more generally, on state higher education budgets (Bhatt, Rork, and Walker 2011). The findings show that as earmarked revenue designated to higher education increases, state funding for higher education from the general fund decreases. The authors argue this is likely due to supplanting of general fund appropriations to higher education with newly earmarked revenue sources in order to stabilize total spending while freeing up general fund money for other purposes (Bhatt, Rork, and Walker 2011). Essentially, although lottery funds are intendedly supplementary, state lawmakers make decisions in the context of scarcity, realizing that allocating resources to one arena of state policy limits the ability to fund other programs, which introduces the potential to supplant funding. As a result, when new revenue emerges from lottery earmarks, state lawmakers might be incentivized to fund general education appropriations less, which could free up funds for other programs or tax cuts. Furthermore, this behavior might be politically advantageous for state officials looking to remain palatable to the public; officials can sponsor lotteries and designate the revenues to universally popular policy arenas (such as education) in lieu of raising taxes, while at the same time freeing up general fund money for other purposes that matter to their constituents (McAuliffe 2006).

Based on these results, it would seem that lottery earmarks could be serving one of two potential purposes. First, lottery earmarks could be supplementing higher education spending by serving as an additional source of revenue that bolsters stable appropriations levels originating from the general fund. Additionally, lottery earmarks could be making up for small decreases in general fund appropriations to higher education with supplementary funding that outweighs the potential decrease in general fund appropriations. If lottery earmarks are being utilized to either supplement stable appropriations or to make up for small decreases in funding from the general fund, we interpret this relationship as supplementary despite the potential for fungibility. Alternatively, lottery earmarks could be serving as a way to stabilize funding for higher education while at the same time drastically reducing the allocation of higher education money originating from the general fund. In opposition to the supplementary case, this supplanting impact would be revealed through a negative and significant or null correlation between lottery earmark revenue and higher education funding. In the supplanting case, decreases in higher education funding originating from the general fund would not be outweighed by lottery earmark revenue designated to higher education budgets.

An alternative conceptualization of a supplementary impact would be a dollar-for-dollar increase in lottery earmark revenue and higher education funding—however, due to data limitations, we are not able to capture this level of nuance. Although it would be ideal to fully capture this relationship by delineating separate variables for higher education appropriations that originate from the general fund as opposed to those that originate from lottery earmark revenue, our data only capture total state appropriations or state grant aid without distinguishing the source. However, as we discuss in the next section, we do gather data on nontax support for a limited number of years, which includes lottery earmark revenue, along with other sources. In our supplementary analysis, we draw on this distinction to isolate, as best as we are able, the impact of lottery earmark implementation on appropriations originating from the general fund by subtracting out nontax support from total appropriations and utilizing the leftover amount as the dependent variable.

4.  Data and Measures

To investigate the impacts of lottery earmark policies designated for higher education, we construct a unique annual panel dataset of all fifty states, for the years 1990–2009. We use the years 1990–2009 for a number of reasons. First, the data gathered after 1990 are much more reliable on our outcomes of interest and the majority of lottery earmark policies for higher education were implemented after 1990. Additionally, this twenty-year period allows us to assess pre- and post-implementation dynamics of lottery earmarks and higher education spending. The key variables in the dataset, as well as their data sources, are summarized in table 2.

Table 2.
Descriptive Statistics of Key Variables (1990—2009)
SourceMeanSDMinMaxSkewN
Outcome Variables
Higher education appropriationsa (in millionsSHEEO 1,298 1,685 44.8 12,836 3.97 1,000
Need-based financial aid NASSGAP 94.4 171.5 1040.9 3.2 1,000
Merit-based financial aid NASSGAP 29.6 76.6 671.6 4.5 1,000
Higher Education Variables
Post*Treatment Constructed 0.31 0.46 0.84 1,000
Post*Scholarship Constructed 0.18 0.38 1.67 1,000
Nontax support (in millionsSHEEO 18.4 75.6 829 5.69 1,000
Economic Controls
Per capita income BEA $34,651$5,941 $21,539$58,499 0.65 1,000
Unemployment rate BLS 5.3% 1.59% 2.3% 13.4% 0.98 1,000
Political Controls
Republican governor KSPD 0.51 0.50 −0.06 1,000
Legislative professionalism Squire 0.19 0.13 0.03 0.66 1.8 1,000
SourceMeanSDMinMaxSkewN
Outcome Variables
Higher education appropriationsa (in millionsSHEEO 1,298 1,685 44.8 12,836 3.97 1,000
Need-based financial aid NASSGAP 94.4 171.5 1040.9 3.2 1,000
Merit-based financial aid NASSGAP 29.6 76.6 671.6 4.5 1,000
Higher Education Variables
Post*Treatment Constructed 0.31 0.46 0.84 1,000
Post*Scholarship Constructed 0.18 0.38 1.67 1,000
Nontax support (in millionsSHEEO 18.4 75.6 829 5.69 1,000
Economic Controls
Per capita income BEA $34,651$5,941 $21,539$58,499 0.65 1,000
Unemployment rate BLS 5.3% 1.59% 2.3% 13.4% 0.98 1,000
Political Controls
Republican governor KSPD 0.51 0.50 −0.06 1,000
Legislative professionalism Squire 0.19 0.13 0.03 0.66 1.8 1,000

Notes: In the empirical analysis, we utilize the natural log transformation of each outcome variable. SHEEO = State Higher Education Officers; BEA = Bureau of Economic Analysis; BLS = Bureau of Labor Statistics; COE = Center for Opportunity in Education, KSPD = Klarner State Politics Dataset; NASSGAP = National Association of State Student Grant and Aid Programs.

aThis variable, as well as the nontax support variable, is calculated in constant dollars to account for inflation and is converted into millions to allow for ease of interpretation in the empirical results.

The data underlying this analysis fall into four main categories: (1) state appropriations data, (2) state financial aid data, (3) lottery earmark implementation data, and (4) time-variant contextual variables. First, we collected data from the State Higher Education Executive Officers Association (SHEEO) on total state appropriations to public universities in each state. The measure of total state appropriations dedicated to higher education is converted into constant 2009 dollars to adjust for inflation using the Higher Education Cost Adjustment index. We also include data on nontax support in constant 2009 dollars in our supplementary analysis because SHEEO incorporates money drawn from lottery and gaming revenue streams in their calculation of total state appropriations.6 Second, we collected data on state need-based and merit-based aid from the National Association of State Student Grant and Aid Programs. These dependent variables reflect our main outcomes of interest: state appropriations for higher education and student financial aid levels.7 Third, we include multiple indicators that capture the implementation of lottery earmarks designated for higher education. Specifically, we constructed a dichotomous measure for all states that designate lottery earmark revenues to support higher education and a separate measure that indicates the years during which each of these state lottery earmark policies have been in place. We created these measures based on state lottery commission reports and based on correspondence with the state lottery commissions, as well as the state higher education governing institutions and state legislative officials. Overall, twenty-five states had lottery earmark policies that contributed to higher education funding during the time period of interest (1990–2009). In addition, we include a dichotomous variable that captures whether some portion of the lottery earmark revenue in each of the states is dedicated to a merit-based or need-based scholarship program. This measure allows us to incorporate a nuanced analysis that accounts for differences in earmark policy structure across states.

Finally, we include data on the time-variant contextual factors that previous research has found to be significant in higher education budgets. Scholars have found that economic and demographic variables have significant impacts on state public higher education appropriations.8 As unemployment increases, states decrease spending on postsecondary education and financial aid, since periods of less-than-ideal economic conditions can make investments in higher education difficult to sustain (McLendon, Hearn, and Mokher 2009). Thus, we control for the unemployment rate of the state as provided by the Bureau of Labor Statistics. On the other hand, increases in per capita income can lead to increased higher education spending (McLendon, Tandberg, and Hillman 2014), which is why we include per capita income measured in constant 2009 dollars as provided by the Bureau of Economic Analysis. Furthermore, to account for the political climates of the individual states, we include a measure of legislative professionalism as originally created by Squire (2007), in which higher scores indicate a more professional legislature. This measure is constructed on a scale from 0.0 to 1.0, where 1.0 represents a legislature that perfectly resembles the U.S. Congress, based on the number of days the legislature meets each year and the salary and staff resources provided to state legislators (Squire 2007). Finally, we include an indicator for states with Republican governors based on the Klarner State Politics Dataset (2013). Once we account for the partisanship of the governor and the level of legislative professionalism, we are able to control for the political influences that have been shown to significantly influence the level and the dynamics of state appropriations over time (McLendon, Hearn, and Moker 2009; Ryu 2011).

5.  Empirical Approach

Our empirical approach seeks to explain the variation in (1) state appropriations to higher education and (2) state financial aid levels as a function of the implementation of lottery earmarks designated to higher education. We utilize a multitude of modeling strategies to estimate the impact of lottery earmarks on higher education appropriations, as well as state need-based and merit-based financial aid. First, we utilize a standard DID modeling strategy, which estimates the impact of a policy by comparing the difference between the pre-intervention and post-intervention changes in treated and untreated states for an outcome of interest. As such, this DID approach leverages plausibly exogenous variation in the adoption of lottery earmarks to address potential selection bias (Riegg 2008). However, it should be noted that in the absence of random assignment, the comparison of states with lottery earmarks to states without lottery earmarks over time will only uncover causal estimates if the intervention is exogenous. This will be explored more fully in the sections laying out the identification assumptions and robustness checks for these estimates.

We implement the DID approach in a regression framework, summarized in equation 1:
$Yit=α+β(Fi*Tt)+θi+λt+βXit+εi,t.$
(1)

In this model, the outcome variables Yit are a function of a constant (α), a state fixed effect (θi), a year fixed effect (λt), a set of covariates (Xi,t), an error term ($ε$i,t), and an interaction between a dichotomous variable for treated states (Fi) and a dichotomous variable indicating post lottery earmark implementation (Tt). The parameter of interest (β) reveals the association between lottery earmark adoption and higher education budgets in post-implementation years for treated states. In addition, we include an alternative specification that includes an interaction between an indicator for states that have lottery earmarks supporting scholarship programs and post-implementation years in order to explore the potential for earmark structure to mediate the findings. We also include state and year fixed effects to account for both time-invariant individual specific factors as well as individual-invariant unobserved time factors. These fixed effects are included because we expect that there are time-specific impacts, such as the financial collapse of 2008, and also state-specific impacts that influence appropriations to higher education. Moreover, we include state-specific time trends in each of the model specifications to account for unobserved heterogeneity not captured in the state and year fixed effects. Additionally, we cluster the standard errors at the state level in accordance with the best practices outlined by Bertrand, Duflo, and Mullainathan (2004).

In the estimation of this model we utilize multiple different comparison groups in order to increase confidence in the results. First, we estimated the model with all nonadopting states as the comparison group. Second, we estimate this model with future adopters as the comparison group in case there are systematic differences between lottery earmark adopters and nonadopters. To explore this possibility and investigate the appropriateness of utilizing all nonadopting states as the comparison group, we provide visual representations of the annual averages in figures 13 for these two groups over the time period. Although it would be ideal to center these figures on the year of implementation and display the comparison of treated and untreated state averages in pre- and post-implementation years, this is not possible given the variation in years of implementation in our data.9 However, these figures do provide a visualization of whether these two groups of states are comparable in changes over the time period. It is encouraging to see that in the earlier years (pre-2000), when many states have yet to adopt lottery earmarks, the averages in the outcome variables are similar, and only begin to diverge in later time periods as more adopting states move into the post-implementation time period. This is especially encouraging given that our calculated mean/median year of implementation is between 1998 and 1999. Thus, this visual evidence suggests that average appropriations and state financial aid levels are similar in pretreatment years. However, this is only suggestive of parallel trends. In our formal analysis, we provide definitive evidence on the parallel trend assumption.

Figure 1.

Figure 1.

Figure 2.

Figure 2.

Figure 3.

Figure 3.

Despite the suggestive evidence in support of the parallel trend assumption when all nonadopting states are utilized as the comparison group, we also specify models with future adopters as the comparison group, in case there are systematic differences between adopters and nonadopters. This sample, in which we utilize future adopters as the comparison group, includes all states that adopted lottery earmarks during our time period (identified as adopters in table 1). This means we exclude the six states that adopted lottery earmarks prior to 1990, seeing as we cannot observe any pre-implementation time periods.

Our second modeling strategy, what we refer to as the flexible DID approach or an event-study model, is a slight modification of the DID model just presented. It is also implemented in a regression framework summarized in equation 2.
$Yit=α+∑n=-66βnDn+βXi,t+θi+λt+εi,t.$
(2)

In this model, our outcome variables (Yit) are estimated as a function of an error term ($ε$i,t), a constant (α), individual state (θi) and year fixed effects (λt), a set of covariates (Xi,t), and a set of dichotomous variables capturing the year relative to implementation for all states that have lottery revenues dedicated to higher education ($∑n=-66βnDn$).10 This model introduces flexibility into the difference-in-difference setup by allowing for the associations between lottery earmarks and higher education funding to vary based on the year relative to implementation. In order to construct this model, we create dichotomous variables to represent ±5 years relative to lottery earmark implementation, as well as a dichotomous variable for six or more years prior to implementation and six or more years post implementation to account for those cases falling outside of the five-year range. We then exclude the dichotomous variable representing one year prior to implementation as the reference group; thus, all comparisons are made to the year prior to implementation. In addition, we include the set of pre-implementation dummies to assess trends prior to implementation.

This model is also commonly referred to as an event-study model (Brown et al. 2016) and is featured in previous literature on the impact of school closures on student achievement (Brummet 2014) and the effects of job displacement on earnings (Jacobson, LaLonde, and Sullivan 1993; Stevens 1997). Similar to previous studies using this specification, we are empirically examining a treatment that contains variation in the timing of implementation and has potential dynamic post-implementation impacts. In this flexible DID modeling approach we are able to both specify the impact of lottery earmarks over the duration of implementation and test the parallel trends assumption explicitly in the model. Although we may observe an immediate level increase in appropriations, it is uncertain whether this will change over the duration of implementation. For example, one can imagine a scenario in which lottery revenues increase and accrue over time, leading to larger positive coefficients on later years. On the other hand, one can imagine a scenario in which lottery revenues represent a one-time increase to higher education appropriations but decrease over time as state legislatures begin to use lottery funds to replace instead of supplement higher education budgets. In this case, we may expect to see positive coefficients that indicate an increase in funding at first but then reflect decreased levels of funding over time. Taken together, given the ability to empirically test an important identification assumption of the DID model, the established use in economic literature, and an ability to examine post-treatment dynamics, we estimate the flexible DID approach as an additional specification to the standard DID model.

6.  Identification Assumptions and Causality

In order to make causal claims in a DID approach, the intervention should be plausibly exogenous and the models must not violate the parallel trends and history assumptions (Bertrand, Duflo, and Mullainathan 2004). In the context of lottery earmark adoption, some of these conditions are not met, which leads us to characterize the estimates as descriptive instead of causal. We begin by discussing the techniques we use to address endogeneity and then address the parallel trends and history assumptions.

First, the most important underlying assumption of our empirical approach is that the timing of lottery earmark adoptions is exogenous. In this context, states may adopt lottery earmark policies for higher education based on decreased spending on higher education in previous years due to declining tax revenue sources or other budgetary pressures. Therefore, it is likely that the adoption of lottery earmarks is endogenous, and changes in spending on higher education after the implementation of these earmarks could be the result of a broader change in policy. For example, one could imagine that after a few years of budget cuts, universities lobby the state legislature to expand support for higher education, which state lawmakers provide through a politically feasible nontax funding source such as lottery earmark revenues. To account for this possibility, we include a dynamic panel model to account for the possibility that our previous DID fixed effects models have not accounted for this unobserved heterogeneity.

Specifically, we implement an Arellano-Bond robust estimator to account for the possibility that state lawmakers consider previous years of funding when making funding decisions (Arellano and Bond 1991). This model combines a first-differences approach described in equation 3 with an instrumental variables technique:
$ΔYit=∅ΔYit-1+δΔEit+Δλt+ΔβXit+Δεit.$
(3)

If we assume inputs are strictly exogenous and that there is no serial correlation in $ε$it, it is possible to utilize the lagged values of $ε$it and Xit as instrumental variables. This method builds on the first-difference model presented in equation 3 by accounting for the correlation between ∆Yit−1 and ∆$ε$it. In this way, the Arellano-Bond robust estimator provides a dynamic panel model that utilizes an instrumental variables technique in combination with first differencing to account for unobserved heterogeneity, in which the lags of the dependent variable and the independent variables are included as regressors (Arellano and Bond 1991). In each of these models, we include state-specific time trends and conduct robustness checks for serial correlation, which we report in the following sections.11

Next, we explicitly test the parallel trend assumption in the flexible DID model specification. In this context, to meet the parallel trend assumption, we need evidence that the pretreatment trends in state appropriations and state financial aid levels are similar for the treatment and comparison groups. This means that trends in the outcomes of interest in the years prior to implementation should be similar for adopters and nonadopters as well as for adopters and future adopters (our robust comparison group). In our modeling framework, one can assess the parallel trend assumption by examining the coefficients on the pretreatment interaction coefficients. First and most importantly, no recognizable pattern should emerge from these interaction coefficients. Additionally, these coefficients will ideally be statistically insignificantly different from zero.

Finally, the DID approach relies on the assumption that any change in the outcome is attributable to the adoption of lottery earmarks and not another simultaneous policy intervention. We recognize the history may be violated in this context. It is possible that states are enacting simultaneously occurring reforms to higher education funding in the same year that lottery earmarks are adopted. For instance, in the same year that lottery revenue is earmarked to higher education, states could also be considering a multitude of other nontax revenue sources, such as casino gaming revenue, to increase higher education funding or offset decreases in previous years. This potential simultaneity of policy reforms impacting higher education funding makes it difficult, if not impossible, to make credible causal claims.

Furthermore, given the immense variation in the duration of implementation, policy designs, and reforms over time across states, it is unlikely that we could reasonably account for all possible simultaneous reforms and for all reforms over time (Novarro 2005). Therefore, due to the possible violation of the history assumption, we encourage cautious interpretation of the estimates provided. Extrapolating a causal relationship would be misguided given the potential for simultaneous reforms that impact higher education funding. Although we recognize the serious challenges to causal inference and empirical identification, we note that many of the most important public policy questions face these challenges. Furthermore, we argue that despite the difficulty faced in identifying causal estimates in substantively important policy questions, these questions are still worth investigating. For this reason, we present this descriptive analysis as a first step toward better understanding the significant and largely unexplored question of whether lottery earmarks supplement or supplant higher education funding.

7.  Results

Lottery Earmarks and Higher Education Appropriations

The results of this analysis, summarized in table 3, suggest that the implementation of lottery earmarks is positively associated with state appropriations to higher education. In the first two columns, the DID models reveal a positive and significant increase in appropriations of approximately 5 percent in post-implementation years.12 Additionally, the second DID specification results suggest that states with lottery earmarks structured to support scholarships are not significantly different from those without structured scholarship support.

Table 3.
Estimates of the Effects of Lottery Earmarks on Appropriations, Future Adopters as Controls
Natural Log of Appropriations
DIDDIDArellano-BondArellano-Bond
Post × Treatment 0.053*** 0.056** 0.003 0.025
(0.013) (0.024) (0.019) (0.029)
Post × Treatment × Scholarship  −0.004  −0.029
(0.029)  (0.031)
N  380 323 323
R2  0.549
F statistic 16.854*** (df = 23; 319) 16.203*** (df = 24; 318)
Wald statistic   1831.1 1863.89
Natural Log of Appropriations
DIDDIDArellano-BondArellano-Bond
Post × Treatment 0.053*** 0.056** 0.003 0.025
(0.013) (0.024) (0.019) (0.029)
Post × Treatment × Scholarship  −0.004  −0.029
(0.029)  (0.031)
N  380 323 323
R2  0.549
F statistic 16.854*** (df = 23; 319) 16.203*** (df = 24; 318)
Wald statistic   1831.1 1863.89

Notes: Models include covariates, two-way fixed effects, and state-specific time trends. Robust standard errors clustered by state in parentheses. DID = difference-in-differences.

**p < 0.05; ***p < 0.01.

In the flexible DID model, summarized in figure 4, the post-implementation interactions suggest a significant increase of similar magnitude, between 4 percent and 6 percent, although it varies from year to year. As figure 4 indicates,13 all coefficients for two or more years post lottery earmark implementation are positive and statistically significant, suggesting that lottery funds supplement total appropriations to higher education. Moreover, this model suggests that the parallel trend assumption may not hold; in years prior to lottery earmark implementation, current adopters appear to have slightly lower levels of appropriations. Indeed, the joint significance test of the reimplementation interactions is significantly different from zero (though only at the 10 percent significance level). However, examining figure 4 suggests a relatively flat—neither consistently positive nor negative—pattern among appropriation in years prior to implementation. Our evidence regarding the parallel trends assumption for appropriations is thus mixed. Still, because previous levels of funding could drive the decision to adopt lottery earmarks, we attempt to account for this endogeneity in our Arellano-Bond specifications. The Arellano-Bond estimator reveals a smaller and statistically insignificant increase in appropriations in post-implementation years in the first specification.14 However, once we account for structural differences in the lottery earmarks, the results resemble the DID models, with an approximately 2.5 percent increase in appropriations in post-implementation years. Although this increase is not statistically significant likely because of reduced power in this model, the result is consistent with the other models and represents a substantively meaningful change in appropriations. Together, these models suggest that lottery earmarks supplement, not supplant, state appropriations to higher education.

Figure 4.

Effect of Lottery Earmarks on State Appropriations to Higher Education

Notes: All pretreatment dummies jointly significant at 0.1 level. All post-treatment dummies jointly significant at 0.01 level.

Figure 4.

Effect of Lottery Earmarks on State Appropriations to Higher Education

Notes: All pretreatment dummies jointly significant at 0.1 level. All post-treatment dummies jointly significant at 0.01 level.

Lottery Earmarks and Student Financial Aid

The model results for the impact of lottery earmarks on state financial aid levels are summarized in table 4 for merit-based aid and table 5 for need-based aid. First, our results indicate that the implementation of lottery earmarks for higher education is associated with substantial increases in merit-based financial aid. The first DID specification reveals an approximately 135 percent increase in merit-based financial aid in post-implementation years. This substantially positive impact is also reflected in the DID specification including the interaction accounting for lottery earmark policies that designate revenue to scholarship programs. In this specification, the increase in merit-based aid is mediated by the structure of the lottery earmark. In states in which lottery earmarks are dedicated specifically to scholarship programs, there is a large and significant increase in merit-based aid, whereas the states with lottery earmarks for general higher education purposes do not experience these supplementary impacts.

Table 4.
Estimates of the Effects of Lottery Earmarks on Merit-Based Aid, Future Adopters as Controls
Natural Log of Merit-Based Aid
DIDDIDArellano-BondArellano-Bond
Post × Treatment 0.856*** −0.395*** 0.843*** 0.160
(0.152) (0.131) (0.303) (0.162)
Post × Treatment × Scholarship  1.569***  0.844**
(0.206)  (0.389)
N 380 380 323 323
R2 0.639 0.670
F statistic 24.524*** (df = 23; 319) 26.959*** (df = 24; 318)
Wald statistic   115.81 40.59
Natural Log of Merit-Based Aid
DIDDIDArellano-BondArellano-Bond
Post × Treatment 0.856*** −0.395*** 0.843*** 0.160
(0.152) (0.131) (0.303) (0.162)
Post × Treatment × Scholarship  1.569***  0.844**
(0.206)  (0.389)
N 380 380 323 323
R2 0.639 0.670
F statistic 24.524*** (df = 23; 319) 26.959*** (df = 24; 318)
Wald statistic   115.81 40.59

Notes: Models include covariates, two-way fixed effects, and state-specific time trends. Robust standard errors clustered by state in parentheses. DID = difference-in-differences.

**p < 0.05; ***p < 0.01.

Table 5.
Estimates of the Effects of Lottery Earmarks on Need-Based Aid, Future Adopters as Controls
Natural Log of Need-Based Aid
DIDDIDArellano-BondArellano-Bond
Post × Treatment −0.126** −0.263*** −0.079 −0.077
(0.051) (0.099) (0.064) (0.102)
Post × Treatment × Scholarship  0.172  −0.222
(0.118)  (0.160)
N 380 380 323 323
R2 0.668 0.670
F statistic 27.939*** (df = 23; 319) 26.907*** (df = 24; 318)
Wald statistic   1027.45 737.99
Natural Log of Need-Based Aid
DIDDIDArellano-BondArellano-Bond
Post × Treatment −0.126** −0.263*** −0.079 −0.077
(0.051) (0.099) (0.064) (0.102)
Post × Treatment × Scholarship  0.172  −0.222
(0.118)  (0.160)
N 380 380 323 323
R2 0.668 0.670
F statistic 27.939*** (df = 23; 319) 26.907*** (df = 24; 318)
Wald statistic   1027.45 737.99

Notes: Models include covariates, two-way fixed effects, and state-specific time trends. Robust standard errors clustered by state in parentheses. DID = difference-in-differences.

**p < 0.05; ***p < 0.01.

Further, the flexible DID model, summarized in figure 5, reveals that the implementation of lottery earmarks for higher education increases state merit-based aid levels immediately after implementation and consistently over the duration of implementation. Figure 5 also suggests an underlying negative trend in merit-based aid precedes the implementation of lottery earmarks to higher education. Given this negative sloping pretreatment trend, it is possible that our estimates understate the effect of lottery earmarks on merit-based financial aid. A joint significance test on the pretreatment dummies also suggests a non-zero pretreatment trend. However, the results from the Arellano-Bond specifications are remarkably consistent with the estimates from the DID models, which increases our confidence in the estimates. Indeed, these models reveal substantively similar increases in merit-based financial aid in post-implementation years (∼132 percent). Together, these findings consistently reveal the substantively large and significant supplementary impact of lottery earmarks on merit-based aid.

Figure 5.

Effect of Lottery Earmarks on State Appropriations to Merit-Based Aid

Notes: All pretreatment dummies jointly significant at 0.1 level. All post-treatment dummies jointly significant at 0.001 level.

Figure 5.

Effect of Lottery Earmarks on State Appropriations to Merit-Based Aid

Notes: All pretreatment dummies jointly significant at 0.1 level. All post-treatment dummies jointly significant at 0.001 level.

In contrast, we find lottery earmarks to be associated with decreases in need-based financial aid levels. The first DID specification reveals a 12 percent decrease in need-based financial aid levels in post-implementation years. Next, in the DID model, which includes the scholarship interaction, the negative association is again mediated by whether the lottery earmark allocates revenue to scholarships. In this case, it appears that states without lottery scholarship programs face significant decreases in need-based aid following the implementation of lottery earmarks, whereas there is a null association for states with lottery funded scholarships. At first glance, this could suggest that higher education funding is zero-sum and those states investing more in appropriations are supplanting need-based financial aid. However, this should be interpreted with caution given the imprecision of the estimates and the opposite, although statistically insignificant, associations found in the Arellano-Bond specifications with the scholarship interaction variable.

Furthermore, this negative association is consistent in the flexible DID specification. In fact, this model reveals that the negative association appears to increase over time. For instance, in the first two years following lottery earmark implementation, the impacts hover around 8 to 10 percent decreases, but by the fifth year of implementation, the decrease jumps to 33 percent. However, as table 5 and figure 6 demonstrate, the pre-implementation interactions are jointly significant, suggesting potential violation of the parallel trend assumption. In fact, in figure 6, it appears that there is a downward sloping pretreatment trend, suggesting these estimates might overstate the negative effect of lottery earmarks on need-based financial aid. Nevertheless, our second Arellano-Bond specification, which accounts for the potential endogeneity of the treatment and the negative trend, produces estimates that are remarkably consistent with those in the DID. Across all models, lottery earmarks are consistently associated with substantial decreases in need-based aid.

Figure 6.

Effect of Lottery Earmarks on State Appropriations to Need-Based Aid

Notes: All pretreatment dummies jointly significant at 0.01 level. All post-treatment dummies jointly significant at 0.05 level.

Figure 6.

Effect of Lottery Earmarks on State Appropriations to Need-Based Aid

Notes: All pretreatment dummies jointly significant at 0.01 level. All post-treatment dummies jointly significant at 0.05 level.

8.  Supplementary Analysis and Robustness Checks

To further investigate whether intendedly supplementary funds are being used to supplant general appropriations to higher education, we perform a supplementary analysis on the impact of lottery earmark implementation on total state appropriations less the nontax support for the years in which data are available (2000–2009). As such, this analysis factors out the funding from lottery, gaming, and tobacco settlements for higher education to account for state legislators potentially supplanting general appropriations with funding from these sources. As figure 7 and table 6 show,15 the implementation of lottery earmark policies for higher education still appears to be positively associated with state appropriations. When compared to the models in our main analysis on total appropriations to higher education, appropriations minus nontax support is supplemented by a similar though slightly smaller increase in appropriations. These results enhance the robustness of our findings that lottery earmarks are associated with supplementing higher education appropriations.16

Figure 7.

Effect of Lottery Earmarks on State Appropriations to Higher Education Minus Nontax Revenue

Notes: All pretreatment dummies jointly significant at 0.05 level. All post-treatment dummies jointly significant at 0.1 level.

Figure 7.

Effect of Lottery Earmarks on State Appropriations to Higher Education Minus Nontax Revenue

Notes: All pretreatment dummies jointly significant at 0.05 level. All post-treatment dummies jointly significant at 0.1 level.

Table 6.
Estimates of the Effects of Lottery Earmarks on Appropriations Minus Nontax Revenue, Future Adopters as Controls
Natural Log of Appropriations Minus Nontax Revenue
DIDDID
Post × Treatment 0.042*** 0.063*
(0.015) (0.033)
Post × Treatment × Scholarship  −0.027
(0.038)
N 380 380
R2 0.438 0.439
F statistic 10.805*** (df = 23; 319) 10.360*** (df = 24; 318)
Natural Log of Appropriations Minus Nontax Revenue
DIDDID
Post × Treatment 0.042*** 0.063*
(0.015) (0.033)
Post × Treatment × Scholarship  −0.027
(0.038)
N 380 380
R2 0.438 0.439
F statistic 10.805*** (df = 23; 319) 10.360*** (df = 24; 318)

Notes: Models include covariates, two-way fixed effects, and state-specific time trends. Robust standard errors clustered by state in parentheses. DID = difference-in-differences.

*p < 0.1; ***p < 0.01.

Furthermore, we conduct a series of robustness checks to increase confidence in the estimates presented in the results section above. First, we supplement our main specifications with a placebo test for each of the outcomes of interest. To construct the placebo, we created a new placebo treatment variable in which states that adopted lottery earmarks sometime during our time period of interest are considered treated in the five years prior to implementation.17 Then, we plug this placebo treatment variable into the identical DID model presented earlier. The results of these placebo tests are all null, which provides additional evidence in support of our findings.

Furthermore, we rerun the models excluding all covariates under the concern that our covariates represent possible downstream effects of the policy. In most cases, such as that of per capita income, it seems plausible that increased funding for higher education following the implementation of lottery earmarks could influence the income levels of the citizenry. However, the models excluding the covariates are remarkably similar to specifications in which covariates are included. For example, the associations with each of the outcomes of interest are almost identical in direction and magnitude to the models presented in the main analysis.18 Therefore, we conclude that potential downstream impacts on our covariates are not problematic for the design.

Next, we include the results with all states as the comparison group instead of future adopters, which are displayed in the tables in the online Appendix. Again, the results are consistent across these different specifications. We also conducted all analyses with all adopters as opposed to only future adopters, thus including the first six states in table 1.19 These analyses were remarkably similar in magnitude, direction, and significance. Finally, we conducted the analysis excluding Florida and Georgia in case these large states could be driving the results. These models are also consistent with our main models with some slight statistical significance changes; however, policy implementations remain statistically significant.20 Finally, we constructed dummies representing years up to eight years prior to implementation. The figures from these models are in the online Appendix. The results of these models are similar to our main findings. These models and the associated figures demonstrate further confidence that we do not violate the parallel pre-trend assumption for appropriations. In fact, pre-treatment dummies are no longer jointly statistically significant. Figures A.3 and A.4 in the online Appendix also more clearly demonstrate the violation of the parallel pre-treatment assumption for need-based and merit-based financial aid, which leads us to interpret these estimates with caution.

Across all specifications of our model, we consistently find that the implementation of lottery earmarks has a lasting and positive association with general appropriations to higher education and state-based merit-based aid. Additionally, we find lottery earmarks are associated with a negative impact on state need-based aid. Specifically, lottery earmarks are associated with an approximately 5 percent increase, relative to the first year before implementation, in appropriations to higher education. For merit-based aid, lottery earmarks are associated with an increase, relative to the first year before implementation, of approximately 135 percent. Finally, we find that lottery earmarks are associated with an estimated 12 percent decrease in need-based aid. These results are in line with prior research that finds there is a trade-off between need-based aid and merit-based aid: when one increases (in this case merit-based aid), the other (need-based aid) decreases (McLendon, Tandberg, and Hillman 2014). In the following section, we discuss the policy implications of these findings for state policy makers as well as public higher education institutions.

9.  Discussion and Conclusion

This article investigates one of the most ubiquitous trends in state higher education budgetary policy: the implementation of intendedly supplementary state lottery earmark funds. By 2009, twenty-five states utilized lottery earmark policies with the intention of supplementing state higher education budgets through contributions to financial aid and general appropriations. However, despite the supplementary intent, scholars have questioned whether designating lottery earmark funds leads to unintended supplanting of higher education funding when state legislators face zero-sum budgetary decisions (Dye and McGuire 1992; Miller and Pierce 1997; Bhatt, Rork, and Walker 2011; Covert 2014).

The implications of these budgetary policy decisions are especially relevant considering the troubling trends in higher education finance. Although educational attainment is required for the economic and social well-being of individuals and society, state budgets have not kept up with increasing demand by students (Perna and Finney 2014). Furthermore, economic turmoil often leads to dwindling state appropriations for higher education, which is associated with increasing tuition at public universities whose students bear the cost of these increases (Weisbrod, Ballou, and Asch 2010). Thus, with students and families facing tuition increases and state funding shortfalls, the impact of utilizing lottery earmark funds for higher education directly impacts the affordability of public institutions, which has far-reaching policy implications for access, retention, and graduation (Weisbrod, Ballou, and Asch 2010; Perna and Finney 2014).

To measure the effect of lottery earmark funds on higher education, this paper exploits temporal variation in the implementation of designated lottery earmarks to higher education on the two outcomes of interest: state appropriations to higher education and financial aid levels. The results of this analysis suggest that, as intended, the use of lottery earmarks is associated with increases in state appropriations to higher education. These results hold even when we subtract out the revenue that was obtained via nontax support, which includes lottery earmarks, suggesting lottery funds do not have unintended supplanting impacts on state appropriations to higher education. Turning to our results for financial aid, which are descriptive in nature, we find strongly consistent estimates of interest. Specifically, the designation of lottery funds for higher education is associated with a substantial increase in merit-based aid; importantly, our models that display the pretreatment trends imply that we may be understating the positive effect we find. In contrast, our estimates indicate that lottery earmarks are negatively associated with need-based aid levels; however, in this case, our models may overstate the negative effect we find.

At a time when many students and families are struggling to pay for college, increasing higher education appropriations and student financial aid could be a key to improving the access, retention, and graduation rates, especially for economically disadvantaged students. Whereas need-based aid encourages those least likely to attend college to both enroll and retain from year to year, merit-based aid is awarded to advantaged students who are more likely to be able to attend and pay for college, which exacerbates the racial and economic disparities in access, retention, and completion (Dynarski 2000; Heller and Marin 2001). Therefore, the potential displacement of need-based financial aid has serious distributional consequences that should be considered when state leaders designate lottery earmark funds to appropriations and state student aid.

It should also be noted that our data and models are limited in the ability to capture causal estimates and structural changes in earmarks over time. Specifically, we are unable to uncover plausibly causal estimates due to the issues with endogeneity and potential simultaneous reforms. It is difficult, if not impossible, to identify causal estimates in this context due to the likelihood of state officials factoring in previous years of funding levels in the decision to adopt earmarks, as well as the potential for simultaneous reforms to occur. Also, we are unable to capture an in-depth examination of the impact of structural variation across states and changes made to the lottery earmarks after the first year of adoption, which could be influencing the outcomes in some of the post-implementation years. One could imagine, for instance, that state governments could implement a supplement-not-supplant provision in post-implementation years, which could significantly change the impact on higher education funding. However, we also highlight that this study is unique in its examination of a set of policies across states and time instead of the usual investigation into one state in limited time periods. Although this approach makes it difficult to identify causal effects, it also sets the stage for future studies to investigate individual state lottery earmarks and the impact on higher education funding.

In the future, scholars should continue this line of work by investigating the heterogeneity in structure across states and link that variation to the impact of the earmarks on higher education funding. Future research should also consider the within-state impacts of lottery funding to further clarify the distributional funding implications of designating lottery earmark funds to higher education. To do so, future research would utilize institution-level data to measure whether the impacts of lottery funding on appropriations vary by type of institution within states. It is possible that lottery funds supplement flagships more than other universities, and this could further vary by the structure of the higher education governing board.

Notes

1.

Need-based aid eligibility is based on financial need, and merit-based aid eligibility is purely based on academic merit regardless of financial need (McBain 2011).

2.

According to recent research, state funding makes up 21 percent of public university budgets, on average, across states (Pew Charitable Trusts 2015).

3.

States have adopted programs that leverage lottery revenue for higher education spending since 1971, when New Jersey adopted its state lottery and dedicated the proceeds to K–12 and higher education.

4.

This table describes the designation of lottery revenue to higher education in the first year of implementation, which does not account for changes over time. Though we note the years in which changes were made to the lottery earmark policies, we are interested in the impact of the initial policy change rather than the structural changes over time. However, changes over time in the structure of lottery earmark policies is a major limitation of this analysis (Novarro 2005). We do account for structural differences between states that allocate lottery revenue to scholarship programs in the empirical analysis, but we also recognize that there are other structural differences that we are not able to account for.

5.

Historically, states have provided a far greater amount of assistance to postsecondary institutions and students—65 percent more than the federal government on average from 1987 to 2012; “the states provide over four dollars of support for higher education expenses for every dollar of federal subsidy” (Archibald and Feldman 2006, p. 618).

6.

Nontax support is defined as the money allocated to public higher education institutions from state lottery and gaming revenues and tobacco settlements.

7.

In the analysis, we utilize the natural log of these variables to improve the ease of interpretation.

8.

We also run the models without any covariates and find remarkable similarity in the results. These results are available upon request.

9.

Because of the variation in the year of implementation across states, each of the nonadopting states would be simultaneously considered in the pre- and post-implementation time period depending on the states in question. Therefore, these figures would have to be created separately for each state. This is not possible due to the varying years of implementation, which makes each untreated state average simultaneously apply to both pre- and post-implementation for any given state implementation condition.

10.

In the estimation of this model, we also include state-specific time trends to account for unobserved heterogeneity not captured by the state and year fixed effects.

11.

We specify the models both with and without the homoscedasticity assumption. Due to the remarkable similarity of these results, we present the models that assume constant error variance.

12.

All models reported in this section were tested for autocorrelation in the residual term using the Durbin-Watson test. No significant results were found, suggesting that serial autocorrelation is not an issue in our models.

13.

Coefficient magnitudes as well as model statistics for all flexible DID models represented in figures 35 are reported in tables A.1–A.3, respectively, in a separate online Appendix, which can be accessed on Education Finance and Policy’s Web site at https://www.mitpressjournals.org/doi/suppl/10.1162/edfp_a_00262.

14.

In each Arellano-Bond model in this section, we test for autocorrelation and fail to reject the null hypothesis at the p < 0.05 level, which suggests that we do not violate this assumption in each of the models presented in this section.

15.

Table A.4 in the online Appendix displays coefficient magnitudes as well as model statistics.

16.

These results should be interpreted with caution for multiple reasons. First, we only have data for nontax appropriations to higher education for ten years, 2000–2009, so our conclusions are based on a smaller time period than in the main analysis. Moreover, we are unable to measure how much of the nontax support is from lotteries specifically. However, lottery earmarks are by definition a subset of nontax support, therefore, our findings represent a conservative estimate of the supplementary impacts of lottery earmarks for these transformed data.

17.

This means states that either never adopted lottery earmarks for higher education and states that adopted lottery earmarks for higher education before 1990 were marked missing and excluded from this analysis. Given the systematic differences between adopters and nonadopters revealed earlier, this comparison within states that adopted lottery earmarks across pretreatment time periods provided the optimal source of variation for a placebo test. The reason for the small N-size compared with the main specifications is due to the exclusion of post-treatment years.

18.

These results are available upon request.

19.

Results also available upon request.

20.

Again, these results are available upon request.

Acknowledgments

We would like to recognize Deven Carlson and Samuel Workman for providing insight and expertise that greatly contributed to the completion of this project. In addition, we would like to thank the anonymous reviewers as well as the associate editor, Stephanie Cellini, for their thoughtful comments that led to the substantial improvement of this paper.

REFERENCES

Archibald
,
Robert
, and
David
Feldman
.
2006
.
State higher education spending and the tax revolt
.
Journal of Higher Education
77
(
4
):
618
644
.
Arellano
,
Manuel
, and
Stephen
Bond
.
1991
.
Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations
.
Review of Economic Studies
58
(
2
):
277
297
.
Arnone
,
Michael
.
2003
.
States once again look to lotteries for scholarship dollars
.
Chronicle of Higher Education
49
(
37
):
22
.
Barrow
,
Lisa
, and
Cecilia Elena
Rouse
.
2004
.
Using market valuation to assess public school spending
.
Journal of Public Economics
88
(
9
):
1747
1769
.
Bertrand
,
Marianne
,
Esther
Duflo
, and
Sendhil
Mullainathan
.
2004
.
How much should we trust differences-in-differences estimates
?
Quarterly Journal of Economics
119
(
1
):
249
275
.
Bhatt
,
Rachana
,
Jonathan C.
Rork
, and
Mary Beth
Walker
.
2011
.
Earmarking and the business cycle: The case of state spending on higher education
.
Regional Science and Urban Economics
41
(
4
):
352
359
.
Borg
,
Mary O.
, and
Paul M.
Mason
.
1990
.
Earmarked lottery revenues: Positive windfalls or concealed redistribution mechanisms
?
Journal of Education Finance
15
(
3
):
289
301
.
,
Kevin
, and
John C.
Pijanowski
.
2007
.
Maximizing state lottery dollars for public education: An analysis of current state lottery models
.
Journal of Educational Research & Policy Studies
7
(
2
):
20
37
.
Brown
,
Meta
,
John
Grigsby
,
Wilbert
van derKlaauw
,
Jaya
Wen
, and
Basit
Zafar
.
2016
.
Financial education and the debt behavior of the young
.
Review of Financial Studies
29
(
9
):
2490
2522
.
Brummet
,
Quentin
.
2014
.
The effects of school closings on student achievement
.
Journal of Public Economics
119
:
108
124
.
Cheslock
,
John J.
, and
Matt
Gianneschi
.
2008
.
Replacing state appropriations with alternative revenue sources: The case of voluntary support
.
Journal of Higher Education
79
(
2
):
208
229
.
Cornwell
,
Christopher
,
David B.
Mustard
, and
Deepa J.
Sridhar
.
2006
.
The enrollment impacts of merit-based financial aid: Evidence from Georgia's HOPE program
.
Journal of Labor Economics
24
(
4
):
761
786
.
Covert
,
Bryce
.
2014
.
Why state lotteries never live up to their promises
.
Available
https://thinkprogress.org/why-state-lotteries-never-live-up-to-their-promises-1f911fae58dc/.
Accessed 28 June 2019
.
Dee
,
Thomas S.
2004
.
Lotteries, litigation, and education finance
.
Southern Economic Journal
70
(
3
):
584
599
.
Dye
,
Richard F.
, and
Therese J.
McGuire
.
1992
.
The effect of earmarked revenues on the level and composition of expenditures
.
Public Finance Quarterly
20
(
4
):
543
556
.
Dynarski
,
Susan
.
2000
.
Hope for whom? Financial aid for the middle class and its impact on college attendance
.
NBER Working Paper No. 7756
.
Dynarski
,
Susan
.
2002
.
The behavioral and distributional implications of aid for college
.
American Economic Review
92
(
2
):
279
285
.
Erekson
,
O. Homer
,
Kimberly M.
Deshano
,
Glenn
Platt
, and
Andrea L.
Ziegert
.
2002
.
Fungibility of lottery revenues and support of public education
.
Journal of Education Finance
28
(
2
):
301
311
.
Evans
,
William N.
, and
Ping
Zhang
.
2007
.
The impact of earmarked lottery revenue on K-12 educational expenditures
.
Education Finance and Policy
2
(
1
):
40
73
.
Gordon
,
Nora
.
2004
.
Do federal grants boost school spending? Evidence from Title I
.
Journal of Public Economic
88
(
9
):
1771
1792
.
Hanushek
,
Erik A.
,
Charles Ka Yui
Leung
, and
Kuzey
Yilmaz
.
2014
.
Borrowing constraints, college aid, and intergenerational mobility
.
Journal of Human Capital
8
(
1
):
1
41
.
Heller
,
Donald E.
, and
Patricia
Marin
.
2001
.
Who should we help? The negative social consequences of merit scholarships
.
Cambridge, MA
:
Harvard Civil Rights Project
.
Hines
,
James R.
, and
Richard H.
Thaler
.
1995
.
Anomalies: The flypaper effects
.
Journal of Economic Perspectives
9
(
4
):
217
226
.
Inman
,
Robert P.
2008
.
The flypaper effect
.
NBER Working Paper No. 14579
.
Jacobson
,
Louis S.
,
Robert J.
LaLonde
, and
Daniel G.
Sullivan
.
1993
.
Earnings losses of displaced workers
.
American Economic Review
83
(
4
):
685
709
.
Klarner
,
Carl
.
2013
.
Governors dataset
.
Available
https://dataverse.harvard.edu/dataset.xhtml?persistentId=hdl:1902.1/20408.
Accessed 28 June 2019
.
Long
,
Bridget T.
2004
.
How do financial aid policies affect colleges? The institutional impact of the Georgia HOPE scholarship
.
Journal of Human Resources
39
(
4
):
1045
1066
.
Martin
,
Robert
, and
Andrew
Gillen
.
2011
.
Measuring college affordability
.
Available
http://dx.doi.org/10.2139/ssrn.1734914.
Accessed 28 June 2019
.
McAuliffe
,
Elizabeth
.
2006
.
The state-sponsored lottery: A failure of policy and ethics
.
Public Integrity
8
(
4
):
367
379
.
McBain
,
Lesley
.
2011
. State need-based and merit-based grant aid: Structural intersections and recent trends.
A Higher Education Policy Brief
.
Washington, DC
:
American Association of State Colleges and Universities
.
McLendon
,
Michael K.
,
James C.
Hearn
, and
Christine G.
Moker
.
2009
.
Partisans, professionals, and power: The role of political factors in state higher education funding
.
Journal of Higher Education
80
(
6
):
686
713
.
McLendon
,
Michael K.
,
David
Tandberg
, and
Nicholas W.
Hillman
.
2014
.
Financing college opportunity: Factors influencing state spending on student financial aid and campus appropriations, 1990 through 2010
.
655
(
1
):
143
162
.
Mikesell
,
John L.
, and
C.
Kurt Zorn
.
1986
.
State lotteries as fiscal savior or fiscal fraud: A look at the evidence
.
46
(
4
):
311
320
.
Miller
,
Donald E.
, and
Patrick A.
Pierce
.
1997
.
Lotteries for education: Windfall or hoax?
State and Local Government Review
29
(
1
):
34
42
.
Nguyen-Hoang
,
Phuong
.
2014
.
Tax increment financing and education expenditures: The case of Iowa
.
Education Finance and Policy
9
(
4
):
515
540
.
Novarro
,
Neva K.
2005
.
Earmarked lottery profits: A good bet for education finance
?
Journal of Education Finance
31
(
1
):
23
44
.
Perna
,
Laura
, and
Joni
Finney
.
2014
.
The attainment agenda: State policy leadership in higher education
.
Baltimore, MD
:
Johns Hopkins University Press
.
Pew Charitable
Trusts
.
2015
.
Federal and state funding of higher education: A changing landscape
.
Available
https://www.pewtrusts.org/en/research-and-analysis/issue-briefs/2015/06/federal-and-state-funding-of-higher-education.
Accessed 2 July 2019
.
Riegg
,
Stephanie K.
2008
.
Causal inference and omitted variable bias in financial aid research: Assessing solutions
.
Review of Higher Education
31
(
3
):
329
354
.
Ryu
,
Jay E.
2011
.
Legislative professionalism and budget punctuations in state government sub-functional expenditures
.
Public Budgeting & Finance
88
(
2
):
22
42
.
Shaul
,
Marlene S.
2002
.
Title I funding: Poor children benefit though funding per poor child differs
.
Available
https://www.gao.gov/assets/240/233331.pdf.
Accessed 2 July 2019
.
Spindler
,
Charles J.
1995
.
The lottery and education: Robbing Peter to pay Paul
?
Public Budgeting and Finance
15
(
3
):
54
62
.
Squire
,
Peverill
.
2007
.
Measuring state legislative professionalism: The squire index revisited
.
State Politics & Policy Quarterly
7
(
2
):
211
227
.
Stevens
,
Ann H.
1997
.
Persistent impacts of job displacement: The importance of multiple job losses
.
Journal of Labor Economics
15
(
1
):
165
188
.
Toutkoushian
,
Robert K.
, and
M.
Najeeb Shafiq
.
2009
.
A conceptual analysis of state support for higher education: Appropriations versus need-based financial aid
.
Research in Higher Education
51
(
1
):
40
64
.
Vance
,
Land Q.
, and
Majeed H.
Alsikafi
.
1999
.
A lottery's impact on instructional and noninstructional expenditures and unrestricted revenues for education
.
Journal of Education Finance
25
(
2
):
149
174
.
Weisbrod
,
Burton
,
Jeffery P.
Ballou
, and
Evelyn D.
Asch
.
2010
.
Mission and money: Understanding the university
.
New York
:
Cambridge University Press
.
White
,
Fred C.
, and
Wesley N.
Musser
.
1978
.
Business cycles and state governmental finances: Implications for higher education
.
Higher Education
7
(
2
):
177
192
.
Zumeta
,
William
.
2012
. States and higher education: On their own in a stagnant economy. In
The NEA 2012 almanac of higher education
,
edited by
Mark F.
Smith
, pp.
27
37
.
Washington, DC
:
National Education Association
.