Patterns and Volatility in State Funding for Higher Education, 1951–2006
by Jennifer A. Delaney & William R. Doyle - 2018
Background: Numerous studies have addressed the determinants of higher education appropriations. Extending prior studies that only consider the relationship between higher education and one other state budget category, Delaney and Doyle develop and test an empirical model of the relationship between higher education and all other budget categories. Delaney and Doyle propose that higher education takes the form of a balance wheel in state budgets. They find that higher education is cut more than other budget categories in bad budget years and given larger increases in good budget years. Although previous work advances understanding of how states budget for higher education, it is limited in the length of time considered.
Purpose: This study makes two important contributions to the literature. First, it documents changes in the amount of volatility in state funding for higher education. Second, it identifies patterns in the volatility, and does so over a longer time period than has been investigated in past research, using data that spans over a half century (1951–2006).
Research Design: This study uses a unique panel dataset spanning the period from 1951 to 2006 to quantitatively document changes in the extent of volatility in state funding for higher education. It also identifies and tests for patterns of volatility.
Findings: We find that the level of volatility in state budgeting for higher education has changed over time. We also find evidence of linear (incremental), quadratic (countercyclical), and cubic (balance wheel) patterns of volatility at different points in time.
Recommendations: Our findings indicate that the role of higher education in state budgets is not static and has varied over time. In policy discussions about higher education funding, we think it is important to consider both absolute funding levels and the amount of volatility in funding. We recommend that higher education leaders discuss not only funding levels with their state legislatures, but also discuss volatility in funding patterns. States and higher education have operated under different funding relationships in the past; therefore, it seems possible that policymakers and higher education officials could change their current funding relationship to conform to a pattern that better serves the needs of the state, institutions, and students.
Concern about higher education finance is a perennial issue. In todays economic climate, most states are facing shortfalls in revenues, and most systems of higher education are bracing for cuts. Although concerns about higher education finance are more pronounced during economic downturns, concerns about state ability and willingness to support higher education seem to surface in all types of economic conditions (Hovey, 1999).
Most of the concerns about funding focus on the level of funding (e.g., SHEEO, 2016). However, this paper argues that the predictability of funding also matters to institutions. Each year, state policymakers must make decisions about priorities in order to fund higher education, while institutional leaders often do not know if their budgets will be cut or increased from one year to the next. Volatility seems to consistently plague state funding for higher education. Budget volatility is expected to be a hardship on institutions since it limits long-range planning and causes uncertainty. Cuts in funding often result in unexpected tuition increases as institutions struggle to cope with changes in other funding sources. These increases can be difficult on students and families as they prepare for college. Volatility in funding seems to be of greatest concern for public institutions since these institutions depend on unpredictable state support.
Because of the importance of volatility in state funding for higher education, it is critical for researchers to better understand how states budget for higher education. We are interested in knowing if institutions have always existed in a volatile environment, or if this is a more recent phenomenon. We also determine whether this volatility falls into a distinguishable pattern (or patterns). This study makes two important contributions to the literature. First, it documents changes in the amount of volatility in state funding for higher education. Second, it identifies patterns in the volatility, and does so over a longer time period than has been investigated in past research, using data that spans over a half century (19512006).
Numerous studies have addressed the determinants of higher education appropriations. Clotfelter (1976) finds that income and wages, along with enrollment levels and outmigration, are significant predictors of state spending on higher education. Peterson (1976) also finds that personal income and enrollment levels are correlated with spending for higher education. McLendon, Tandberg, and Hillman (2014) show that wealth is positively associated with state appropriations to institutions. Leslie and Ramey (1986) find that the link between enrollments and appropriations that had existed between 1965 and 1971 had all but disappeared by the late 1970s. Koshal and Koshal (2000) show that state general appropriations for higher education are impacted by institutional tuition levels along with majority party control in the legislature, state revenues, the size of the two-year sector, and demand for higher education in a state. Stone (2016) argues that state appropriations are inversely related to federal Pell grant funding.
Kane, Orszag, and Gunter (2003) find that Medicaid spending is negatively related to higher education spending per capita. State-specific regressions reveal a large amount of variability in this relationship. Contrary to Kane et al. (2003), Rizzo (2003) does not find that higher education spending has been crowded out by spending on other budget categories. Rizzo (2003) instead finds that shifting demographic patterns have been associated with lower funding for higher education. Okunade (2001) shows that Medicaid competes with higher education, but corrections spending augments higher education appropriations. Using data from 19691994, Humphreys (2000) shows that income has a positive and significant effect on appropriations to higher education. Humphreys (2000) also finds that during recessionary periods, higher education is likely to be cut more than other budget categories.
Using a fixed effects model with data from 1984 to 2004, McLendon, Hearn, and Mokher (2009) highlight the importance of politics on state support for higher education. They find that partisanship, legislative professionalism, term limits, interest groups, and gubernatorial power influence appropriations levels (McLendon et al., 2009). In similar work, Nicholson-Crotty and Meier (2003) examine the relationship between governance structures and political forces. Likewise, Heck, Lam, and Thomas (2012) find that political culture in a state mediates economic forces insulating higher education appropriations. Partisanship also affects higher education expenditures, as shown in Dar and Lee (2014). McLendon, Tandberg, and Hillmans (2014) exploration of sources of variation in state spending on student aid and appropriations reveals that postsecondary governance structures matter for the level of state appropriations.
Extending prior studies that only considered the relationship between higher education and one other state budget category, Delaney and Doyle (2007, 2011, and 2014) develop and test an empirical model of the relationship between higher education and all other budget categories. Delaney and Doyle (2007, 2011, and 2014) propose that higher education takes the form of a balance wheel in state budgets. They find that higher education is cut more than other budget categories in bad budget years and given larger increases in good budget years. Although this work advances understanding of how states budget for higher education, it is limited in the length of time considered. The current study focuses on long-term patterns in state spending on higher education by considering the time period between 1951 and 2006.
DATA AND METHODOLOGICAL APPROACH
To test the level of volatility in state funding for higher education and to determine if there are predictable patterns in this volatility, we constructed a dataset for this study using publicly available data that allows for cross-state comparative analysis. It is a panel dataset spanning 55 years, between 1951 and 2006.1 The data are identified by state and year. Many changes occurred in higher education during this time, including the rise of community colleges and a rapid increase in both the number and proportion of women attending higher education institutions. The benefits of using reliable, comparable state-level data outweigh the benefit of potential control variables that we could include by greatly truncating the time period of our dataset. As such, this work makes a unique contribution to the field because of the long timeframe of the analysis. In addition, our collection of state-level data related to higher education from the Statistical Abstracts of the United States back to the 1950s has, to our knowledge, never been used in academic scholarship in the field of higher education finance.2
Our data on state tax appropriations for higher education span from 1951 to 2006. These data come from two sources. The data from 1951 to 1959 are from the Statistical Abstracts of the United States, U.S. Census Bureau.3 The data from 1960 to 2006 are from Grapevine, the study of state appropriations for higher education begun by M. M. Chambers and continued by his successors at Illinois State University (Chambers, 1961; Palmer, 2005) and the State Higher Education Executive Officers.
The data on total general expenditures for each state spans from 1951 to 2006. These data all originated with the U.S. Census Bureau, but were compiled from a few different sources. The first is the Statistical Abstracts of the United States, from which we gathered data from 1951 to 1979. Unfortunately, data for 1973 was not reported, so we have missing values in our dataset for this year.4 The data from 1980 to 2004 was made available to us from a direct data request with the U.S. Census Bureau. The data for 2005 and 2006 were collected from the Statistical Abstracts of the United States.
It is very difficult to find data to use as control variables since most of the data we would normally use in this type of analysis were neither collected before the 1970s nor comparable across states. We were, however, able to include three control variables in our dataset related to the economy of each state. These are the total number of employees in a state, total state personal income, and total state population. The variable for the number of employees in a state was collected from the Bureau of Labor Statistics, State and Area Employment, Hours, and Earnings, and appears in the Statistical Abstracts of the United States. The data are not seasonally adjusted, and we use them as a measure similar to unemployment, which previous literature indicates is related to college enrollments, especially in the two-year sector (Betts & McFarland, 1995). We include total state personal income as a measure of the overall prosperity of the state. This variable represents a measure of income received from all sources during the calendar year by residents of each state. These data were collected from the Statistical Abstracts of the United States. To control for the relative size of each state, we include a control variable for total state population, a variable that was also collected from the Statistical Abstracts of the United States.5
We also include, as controls, measures of the political context of the state. First, we include a measure of the party of the state governor, with states with Democratic governors set to 1 (Klarner, 2003). It has been suggested in other studies that the control of the governors office could impact higher education appropriations, and we sought to control for the impact of governors party on changes in funding over time (McLendon et al., 2009; McLendon et al., 2014). Second, we include a measure of divided government, set to 0 if all three branches of government were controlled by the same party, and set to 1 if any of the three branches of government were not controlled by the same party. There is an extensive body of research on the impact of divided government, much of it suggesting that a divided government can result in substantive differences in policy (Fiorina, 1991). Last, we include two measures of partisan balance in state legislatures: the proportion of the upper house of the legislature controlled by Democrats, and the proportion of the lower house of the legislature controlled by Democrats. McLendon et al. (2009) and McLendon et al. (2014) find that partisanship is a key variable that may affect appropriations. Given this finding, we control for partisanship as it may impact patterns in changes in spending over time as well.
Because Alaska and Hawaii did not join the union until 1959, we do not include these territories in our analysis until they became states. Descriptive statistics for the data used in this study are presented in Table 1. The first section of Table 1 shows descriptive statistics for all years. The following sections show descriptive statistics for each decade. On average, across states, there has been remarkable growth in CPI adjusted state tax support for higher education. This support ranges from $3,900,935 in Vermont in 1953 to $9,069,528,000 in California in 2002.
Table 1. Descriptive Statistics, All Years and by Decade
We use the data described above to measure changes in volatility over more than a half century, and to explore whether state funding for higher education has become more volatile over time. We also seek to discover if the volatility falls into a predictable pattern (or patterns). In this section, we describe three identifiable patterns, which we use as conceptual models to test for patterns of funding in our data. Each of the following patterns of volatility describes the relationship between state expenditures in all other budget categories (excluding higher education) and state tax appropriations for higher education. We test against all other budget categories to see if the pattern for higher education is unique or if it falls in line with general state budgeting patterns.
We simplify our conceptualization of state budgets as existing in either good budget years or bad budget years. We define good budget years as years in which state appropriations for higher education and expenditures in all other budget categories are positive. Bad budget years are years in which state appropriations for higher education and expenditures for all other budget categories are negative. (Mixed budget years would be years in which there are contradictory appropriations for higher education and all other state budget categories.) We represent good and bad budget years graphically, with state appropriations for higher education on the y-axis and other state expenditures on the x-axis, in Figure 1a.
Figure 1a. Good and bad budget year
A linear pattern would describe a relationship where funding for all other state budget categories and funding for higher education increases (or decreases) in a similar manner. In good budget years, spending on higher education would increase at approximately the same rate as spending for all other state budget categories. This pattern would perhaps best describe incremental budgeting, where states maintain a similar distribution of funds each year and increase (or decrease) all categories by a similar proportion. Funding in the previous year would be the best predictor for funding in a subsequent year. A linear pattern is illustrated in Figure 1b.
Figure 1b. Linear pattern
This pattern would describe a time when funding for higher education is out of step with the rest of the state budget. Spending for higher education would be cut or increased (at a higher or a lower rate) in a manner that does not match the direction (or rate) of spending in other state budget categories. A quadratic function would indicate countercyclical differences in spending for higher education, depending on how good the budget environment is. We show one possible quadratic form during good budget times in Figure 1c. When times are only slightly good, higher education spending is cut (at a decreasing rate) while other state budget categories are increased. By contrast, if times are very good, spending on higher education is increased at a faster rate than spending on all other budget categories. In this pattern, there is an inflection point at which higher education shifts from being cut in good budget times to being an attractive area in which to spend money. In addition to the direction of the funding cuts or increases, the rate of spending in comparison to other state budget categories becomes important when higher education spending gets out of step with other state budget categories. In this case, the type of budget environment does not drive the shape of the relationship, as cuts or increases could happen in any type of budget environment.
Figure 1c. Quadratic pattern
This pattern describes a particular type of cubic function. Although it is possible to imagine that the pattern of state funding for higher education when compared to expenditures in all other budget categories could fall into any number of cubic forms, Delaney and Doyle (2007, 2011, and 2014) found evidence that state general appropriations for higher education fall into a balance wheel pattern. Therefore, we will also test for evidence of the balance wheel pattern in this work.
Harold Hovey, a longtime state budget analyst and higher education observer, first proposed the balance wheel idea. Hovey argued that when states revenues are low, higher education is an attractive option for heavy cuts. Higher education is an attractive area to cut because it has the ability to collect fees (generally in the form of tuition) for its services, an ability lacking in most other major state spending categories. When states revenues are high, higher education is a politically attractive area to spend money (Hovey, 1999).
In a balance wheel model, appropriations to higher education increase at a faster rate than all other budget categories during good budget times. This indicates a curve in the upper right quadrant that is concave up. In bad budget times, the reverse would be true and we would expect a curve in the lower left quadrant to be concave down. This function is illustrated in Figure 1d. As with the incremental pattern, the direction of spending matters (cuts or increases); likewise, the rate of spending matters, as it does with the countercyclical (quadratic) spending pattern. In addition, the type of budget environment matters to the predicted direction of spending.
Figure 1d. Balance wheel pattern
This S-shaped curve is most parsimoniously described as a particular type of cubic function with the following form:
y = a + bx + cx2 + dx3, where b and d are positive.
However, because the balance wheel is primarily concerned with change over time, the full specification of the balance wheel model uses first differencing to track year-to-year changes in the data. First differencing the data allows us to control for state-level heterogeneity in our panel data, but also means that the interpretation of our results must be in terms of change in both the independent and dependent variables. First differencing is expressed as follows:
∆yst = yst – ys(t-1)
Putting together the basic balance wheel form and the first differencing, we arrive at the following estimating equation, which we use in our tests for the balance wheel pattern:
∆yst = α+ β1∆xst + β2∆xst2 + β3∆xst3 + β4∆zst + β5∆wst + β6∆vst + β6∆ust + ε
y = state appropriations for higher education
s = state
t = year
α = constant
x = state total expenditures for all other budget categories
z = state total number employees
w = state resident population
v = state total personal income
u = a vector of controls for the political characteristics of each state
ε = error term
CREATING TIME PERIODS
Having established the patterns of volatility that we will test for in the data, it is important to consider the units of time in which we examine different patterns. Although we think it is important to understand volatility and patterns of volatility over the entire 55-year period, we also think that it is useful to divide the data into distinct time periods to better identify changes in patterns of volatility. It is an intellectual challenge to determine the best way to divide the time period covered by our data. We ultimately could not determine one best way to divide the data. Therefore, we have divided our data using a number of different methods including decades, business cycles, and political parties.
Decades are easily understood units of time and are long enough to provide a good deal of data for analysis in our models. The drawback of using decades is that decades are arbitrary and do not necessarily capture comparable years. For this measure, we use all the years that fall within each decade. This means that we have complete 10-year spans for the 1960s, 1970s, 1980s, and 1990s, but are somewhat short of our 10-year goal with the 1950s (as we do not have data for 1950 and are one year short) and the 2000s (as our analysis ends in 2006).
Unlike decades, business cycles are non-arbitrary, measurable units of time that capture both good and bad budget years within each business cycle. However, it is possible that business cycles are too closely related to changes in state expenditures and appropriations for higher education, and may introduce selection bias into our analysis. We use data from the National Bureau of Economic Research (NBER) to define each business cycle in our dataset. We defined business cycles as spanning from the start of a recession to the year before the start of the next cycle. We also ran an analysis of patterns in any recessionary period regardless of the business cycle in which the recession took place.
Prior literature has shown that politics matter to the setting of state appropriations for higher education (e.g., McLendon et al., 2009). For this analysis we tracked the political party of the U.S. president and divided the sample by years of Democratic and Republican control. In addition to these broad categories that span the entire analysis period, we ran separate models for each period before there was a change in the party of the U.S. president.
We should note that there were other marked changes happening to higher education during the time period of our study. Most notable is the expansion of the community college sector. While Joliet, the first community college, was founded in 1901, most of the expansion of the public two-year sector took place after 1950. In 19491950, there were 287 public two-year institutions (excluding branch campuses); by 20052006, there were 1,053 public two-year institutions (including branch campuses) (Digest of Education Statistics, 2006). Not only did the number of institutions expand, but so too did enrollment. In public two-year institutions, total enrollment was 117,000 in 1955, but grew to 977,000 by 2005an eightfold increase (Digest of Education Statistics, 2006). This exponential expansion of higher education enrollment was seen in all sectors, with total postsecondary enrollment in degree-granting institutions increasing from 670,000 in 1955 to 2,657,000 in 2005 (Digest of Education Statistics, 2006). The rate of growth in opportunities for women is also a notable development over the last half-century. In 1955, 254,000 women were enrolled in degree-granting institutions, comprising nearly 40% of undergraduates. By 2005, 1,457,000 women were enrolled in college, comprising nearly 55% of undergraduates. The development of community colleges, growth in total enrollments, and expansion of opportunities for women are only a few of the major changes to take place in U.S. higher education since 1951. Within the context of these changes, we think it is important to know how the funding relationship between states and institutions of higher education has changed during this period, and if this relationship falls into identifiable patterns.
In our analysis plan, we seek to describe patterns of volatility in the data. We do this by reporting regression results from fitting three possible patternslinear, quadratic, and cubic (as described above)to the data. We then subdivide the years of the study by decades, recessionary period, business cycles, U.S. presidential political party, and change in U.S. presidential party to test if the patterns have changed over time. We use regression estimates of the impact of state expenditures for higher education in relation to spending for all other budget categories. As discussed in our methods and conceptual framework sections, we estimate three models (linear, quadratic, and cubic) in order to see which polynomial expansion provides the best description of the relationship. Estimates for each model were computed with robust standard errors clustered by state. Because first differencing the data restricts the variance to be within the year-to-year change for each state and controls for unobserved heterogeneity, we do not use state fixed effects in our models. However, we do include year fixed effects in each model to control for common time trends in the data.
Table 2 shows results from models estimated using all of the years in the dataset. As shown in the table, the quadratic specification offers the best description of the pattern of the relationship between state tax appropriations for higher education and state spending in all other budget categories. In the fully specified model in which we include the cubic term (Model 3), the squared term is positive and significant (p < 0.05), indicating a countercyclical relationship. In all of the models, the number of employees per state and the total state personal income are also positive and significant (p < 0.01).
Table 3 shows results for estimates calculated using decades as cut points for the data, reporting only the fully specified cubic model for each decade. With the analysis by decade, we see that there is not a consistent relationship between higher education and all other state budget categories over time, and different patterns are seen in different decades. In the 1950s, we find evidence of a linear relationship with a borderline negative, significant finding on the linear term (p < 0.1). It is interesting that the relationship is negative, indicating that as state expenditures for other state functions increased, state spending for higher education decreased. In the 1970s, we find evidence of a cubic relationship with a negative, significant linear term (p < 0.05) and a positive, significant cubic term (p < 0.01). While this provides evidence of a complex relationship, it does not fit into the balance wheel model due to the negative sign on the linear variable. In the 1990s, we also find evidence of a cubic functional form. This cubic function is different from the 1970s cubic function, with only a borderline positive, significant term on the cubic coefficient (p < 0.1). However complex, this cubic function also does not fall into the balance wheel pattern. In the 2000s, we find evidence of a quadratic functional form with positive, significant coefficients on both the linear and squared terms (p < 0.1 or smaller) indicating that higher education was on a countercyclical trend from other state budget categories.
Table 3. Results of Model Estimation on State Tax Appropriations for Higher Education, by Decade
Interestingly, in the 1960s and the 1980s, no significant pattern is revealed. If, however, the linear and cubic coefficient values had been significant during the 1980s, then there would have been evidence of the balance wheel functional form. During the 1980s, Medicaid expanded rapidly in many states, which complicated the relationship between state appropriations and higher education. There may be little significant association during this time period because increases in state spending were largely a function of states matching federal Medicaid funding (Kane et al., 2003).
Our results for the decade-by-decade analysis indicate there are different patterns of volatility over time. The patterns of volatility range from a relatively simple linear function in the 1950s to more complicated countercyclical and cubic functions in later decades. Each decade had very different relationships between expenditures in other state budget categories and state appropriations for higher education.
RECESSIONS AND BUSINESS CYCLES
We next turn to our results for recessionary periods and business cycles. Table 4 presents results when grouping together years for any recessionary period between 1951 and 2006. We use the NBER definition of recessions and coded recessionary periods from the start of a recession to the year before the start of the next business cycle (peak to trough). Any year with at least one quarter in a recession is counted as a recessionary year. By grouping all of the recessionary years together, we somewhat remove the linear nature of time and consider whether states treat higher education spending differently during economic downturns. In Model 2, Table 4, we find evidence of a positive, significant quadratic relationship (p < 0.01), but this finding is swamped by the inclusion of the cubic term in Model 3. Model 3 indicates that there is no significant pattern in the relationship between state spending on higher education and state spending on other budget categories during recessionary periods. To the extent that state spending for all budget categories changes during economic downturns, state spending for higher education does not appear to be treated differently than other state budget areas during bad economic times. We also find that the total number of employees and total personal income in a state are positive and significant (p < 0.05 or smaller).
Table 4. Results of Model Estimation on State Tax Appropriations for Higher Education, Any Recessionary Period
Next, we consider separate results for each full business cycle between 1951 and 2006, as shown in Table 5. For these analyses we used the NBER definition of business cycles and coded each period as beginning with the first peak and ending in the year before the next peak (peak to peak). We counted any year with a quarter at a peak as being either a starting or ending year. Because the business cycle in 19801981 was so short, we were unable to calculate a first differenced result within this narrow window. Hence, we combined this business cycle with the 19811989 business cycle.
Table 5. Results of Model Estimation on State Tax Appropriations, by Business Cycle
In the first business cycle captured in our data from 19511959, we find evidence of a cubic functional form with negative, significant coefficients on the linear term (p < 0.05) and a positive, significant coefficient on the cubic term (p < 0.05). This shows evidence of a complicated relationship between state spending on higher education and spending on other budget categories. However, because of the signs of the coefficient on the linear term, the cubic functional form does not follow the balance wheel pattern. It is a somewhat surprising result when compared with the finding of a negative linear pattern for the full decade of the 1950s (Table 2, Model 1). This indicates that the decade-long results mask some of the volatility of the relationship, although the negative sign on the linear term is consistent in both tests.
In the business cycle from 19691972, state spending on higher education shows a negative, significant linear relationship with state spending in other budget categories (p < 0.05). During this business cycle, higher education funding decreased while state spending in other budget areas increased.
In the next business cycle, from 19731979, we find evidence of a cubic function with a negative, significant term on the linear coefficient (p < 0.05) and positive, significant terms on the squared and cubed coefficients (p < 0.05 or smaller). While providing evidence of a complex relationship, this pattern does not match the balance wheel pattern due to the negative coefficient on the linear term.
In the last cycle tested in our dataset, between 2001 and 2006, we find evidence of a positive, significant quadratic relationship between state spending on higher education and other budget categories (p < 0.05). This provides evidence that spending on higher education was countercyclical during this business cycle.
We do not find evidence of a significant pattern in funding for the 19571959, 19601968, 19801981 and 19811989, or 19902000 cycles. If the linear and cubic terms had been significant in the 19801981 and 19811989, and the 19902000 business cycles (Models 6 and 7), then the balance wheel model would have been present in the data. In all of these results, it is possible that business cycles themselves might be too closely related to the variance that we are seeking to identify to produce unbiased estimates. We suspect that these estimates reduce the likelihood of finding any results, since we selected years to match with known volatility in the economy. The fact that we found evidence of all three types of patterns during these different business cycles speaks to the strength of these findings and adds insights beyond the results that were presented by decade.
The third approach we used to consider divisions in our data is to examine different political moments in our nations history to see if these time periods reveal information about the patterns of state budgeting for higher education. We chose political measures to divide our sample because we know that state budgeting for higher education is set within a political environment (for a discussion, see McLendon et al., 2009, and Tandberg & Ness, 2011). We also wanted a way to divide our sample that was not based on economic changes, but did represent the possibility of differing approaches to postsecondary funding (perhaps based on political ideology). In order to apply a consistent measure across all states, we used the political party of the U.S. president to denote different political epochs. In doing this, we are not indicating that presidents exert direct control over state funding for higher education; rather, we are using presidential party control to mark different political eras in the country.
Table 6 presents results for any year in which the U.S. president belonged to either the Democratic or Republican Party. By running models on any year in which a Democrat (or Republican) was in the Oval Office, we remove some of the linear nature of time that is seen in our decade-by-decade results. This approach also enables us to test whether there are general trends in how states budget for higher education under different types of presidential regimes. The table presents results for linear, quadratic, and cubic models for each political party, but we will focus our discussion on the fully specified cubic model (Models 3 and 6). We find that the cubic functional form is borderline significant for the years in which there was a Democrat in the White House. In this model (Table 6, Model 3), we observe positive, significant coefficients on the quadratic and linear terms (p < 0.1). Similarly, we find that the cubic functional form is significant for the years in which a Republican served as Commander in Chief. Under the GOP, we find a negative, significant cubic term (p < 0.01) in Table 6, Model 6. It is interesting that we find evidence of cubic patterns under both presidential parties, but neither of the functional forms fits the balance wheel model. As with the other models tested, the logged total number of employees in a state and the total personal income generated in a state were also positive and significant in every model tested (p < 0.05 or smaller). These findings are consistent with prior literature that has found evidence that partisanship matters in state spending on higher education (e.g., McLendon et al., 2009).
Table 6. Results of Model Estimation on State Tax Appropriations for Higher Education, by Presidential Political Party Any Year
Table 7 presents results by each change in the U.S. presidential party during the time period of our analysis. We do not code each presidential term, but instead code each period to represent the time during which one party remained in control of the executive branch. This means that we combine presidents Kennedy and Johnson into one period, combine presidents Nixon and Ford into another, and combine presidents Reagan and G. H. W. Bush into a single period. We also code each partys period as starting with a January inauguration and ending in the year before a president of the opposite political party was inaugurated (in other words, we did not base our coding on election cycles). We were unable to use the first two years of our analysis under Democratic President Truman because the 19511953 time period did not provide enough years for first differencing. In order to keep the table to a manageable size, we only present results for the fully specified cubic model in Table 7.
Table 7. Results of Model Estimation on State Tax Appropriations for Higher Education, by Change in U.S. Presidential Party
The first presidential period, from 19531960, represents President Eisenhowers time in office as a Republican. During this period (Model 1), we find evidence of a linear relationship with a negative, borderline significant coefficient on the linear term. This indicates that higher education funding was cut proportionally while state spending in other budget categories increased.
During the Democratic Carter administration we find evidence of a cubic function (Model 4). This function has a negative, significant linear coefficient (p < 0.05) and positive, significant coefficients for the squared and cubic terms (p < 0.05 or smaller). This indicates a complex relationship between higher education and other state budget categories during the period between 1977 and 1980. However, the pattern uncovered does not fit the balance wheel model.
We find evidence of a quadratic function under G. W. Bushs Republican presidency (Model 7) with a positive, significant coefficient on the squared term (p < 0.05). This indicates that, between 2001 and 2006, state spending on higher education was countercyclical to state spending in other budget areas.
We do not find evidence of a significant functional pattern during the Kennedy and Johnson period (19611968), the Nixon and Ford period (19691976), the Regan and G. H. W. Bush period (19811992), or the Clinton period (19932000). It is interesting that for a majority of the political periods tested we were unable to detect a significant pattern. This may help to explain higher education institutional leaders sense that state budgeting for higher education is unpredictable. It is also interesting that the periods with no predictable pattern are divided among Democratic and Republican regimes.
Had the linear and cubic coefficients been significant under President Clintons Democratic presidency (Model 6), we would have found evidence of the balance wheel functional form. We find it interesting that there is no clear delineation in the higher education spending patterns by political party. We find evidence of a linear pattern under President Eisenhower and a quadratic function under President G. W. Bushboth Republicans. The most complicated relationship is found under a Democrat, President Carter.
CONCLUSIONS AND POLICY IMPLICATIONS
Volatility in state budgeting is a challenge for both states and higher education institutions. In this study, we find that the level of volatility in state budgeting has varied over timeranging from very stable and predictable relationships to extremely volatile relationships. The volatility we document is generally not random, but falls into discernable patterns. We identify instances of a number of different patternslinear/incremental, quadratic, and cubicin data spanning 55 years, indicating that patterns of volatility change over time. We also report a number of periods in which we were not able to identify a significant pattern. From all of these analyses, it seems clear that there never were golden years of higher education funding, and that cuts and volatility appear in every time period tested.
We suspect that other forces, such as changes in state budgets due to Medicaid spending requirements (Kane et al., 2003; Okunade, 2001), K12 spending requirements (Toutkoushian & Hollis, 1998), or higher education institutional expansion (particularly related to the growth of the community college sector), are influential in state budgeting for higher education. In addition, higher education demographic changes likely play an important role in state funding since many states use some type of formula funding based on enrollments, or completions, as has become more common in recent years (Leslie & Ramey, 1986; Rizzo, 2003; Toutkoushian & Hollis, 1998). However, we were only able to identify consistent and comparable data sources for the broader measures that we tested, not the more nuanced measures that may also be important. We think that our broad measures provide a useful indication of patterns of higher education, and 55-year dataset provides a unique perspective on the relationship between state spending on higher education and state spending in other budget areas.
Given prior work in this area (such as Delaney & Doyle, 2007, 2011, and 2014), we were surprised not to find clear evidence of the balance wheel pattern. We found the balance wheel pattern during the 19801989 decade, during the business cycles from 19801981 and 19811989, during the business cycle from 19902000, and under President Clinton. However, none of these patterns yielded significance. Had the cubic terms been significant, we would have also found the balance wheel pattern for all years, during the decade 20002006, during the 20012006 business cycle, and under President G. W. Bush. The lack of significance on the cubic terms might be due to the roughness of the high-level data that we used for this study, whereas prior studies spanned shorter time periods and had access to more fine-grained data.
In policy discussions about higher education funding, we think it is important to consider both absolute funding levels and the amount of volatility in funding. We recommend that higher education leaders discuss not only funding levels with their state legislatures, but also discuss volatility in funding patterns. States and higher education have operated under different funding relationships in the past; therefore, it seems possible that policymakers and higher education officials could change their current funding relationship to conform to a pattern that better serves the needs of the state, institutions, and students. In particular, both legislators and institutional leaders need to recognize that volatility is very often a feature of state funding for higher education. It could be that such volatility affects institutions negatively beyond just the cuts themselvesif institutions are unable to plan, then their ability to respond to either increases or decreases in funding could be impacted.
In a volatile system, it is also important know whether volatility falls into recognizable patterns and whether those patterns change over time. We urge researchers to continue to measure and identify patterns of volatility and to share this information with policymakers. We hope that future researchers will be inspired by our work and will conduct additional analyses with long time horizons. In particular, future work that identifies and incorporates additional control variables as historical measures would add to the literature. Future research in this area should also include studies that investigate why or how the patterns that we identified in this work developed. We think that better understanding the changes in the funding relationship between higher education appropriations and funding for all other state budget categories can help institutions and states to better understand the past and plan for the future.
We would like to thank Mehir Desi for sharing the work that he did cleaning the U.S. Statistical Abstract data with us. We also thank Patricia Yu and Bradley Hemenway for their research assistance on this project.
1. We selected these 55 years for our panel dataset because 1951 is the earliest year to offer over a half century of data for our analysis. Because of the discontinuation of the Statistical Abstracts of the United States data source in 2012, we selected 2006 as the end year of our analysis. We set the end year of our analysis to 2006 to avoid capturing only part of the Great Recession and its recovery, a time period that cannot be fully captured because of the discontinuation of the Statistical Abstracts (and one in which we expect different behavior among states toward higher education). We leave it to future research to address the volatility of higher education funding during the Great Recession and to consider the role of federal stimulus funds in shaping the relationship during this period.
2. A review of historical quantitative higher education finance studies shows that identifying reliable and consistent data to use over an extended time period is a primary challenge in this subfield (e.g., discussions in Goldin & Katz, 1999). While we understand (and have used in related research) other datasets, there is no other collection of state-level data with consistent measures that span back to 1951. Some datasets, such as IPEDS or the Delta Cost Project, contain a wealth of variables that we could use as controls in our analysis, but these data are only available back to 1984. The Higher Education General Information Survey (HEGIS), the federal data system that predated IPEDS, was in use between 1966 and 1985. While there are some variables that are consistent between HEGIS and IPEDS, many variable definitions changed between the two datasets and the sample of institutions is different. Because of this, a longitudinal dataset comprised of HEGIS and IPEDS raises concerns about data reliability and consistency. In addition, HEGIS and IPEDS do not stretch back into the 1950s. The Digest of Education Statistics only reaches back to 1990. The National Science Foundations Computer Aided Science Policy Analysis and Research Database System (CASPAR) draws its education data from the U.S. Department of Education datasets, and therefore has the same date limitations. The Grapevine survey, and later the State Higher Education Finance report from the State Higher Education Executive Officers, offers longitudinal data on state tax support for higher education back to the 1960s. While a consistent data source, it does not span the full number of years of our analysis and does not include state-level information beyond tax expenditures. We are not aware of any state that has longitudinal data on their higher education systems that reaches back to the 1950s, and we know how challenging it can be to identify comparable measures across state data systems (for a discussion of this issue, see Ewell, 2009; Ewell & LOrange, 2009; and Ewell, Schild, & Paulson, 2003).
3. These years predate the Grapevine survey, which is why a different data source was used.
4. We have remarkably few missing data problems given the long time series in our dataset, but there are two missing data issues worth mentioning. Alaska and Hawaii not being members of the union yields some missing data. In addition, the Statistical Abstracts were not collected in 1973, so data is missing for every state for this year. In the years 19511959, there are 46 states that should have data for eight years. Nebraska is excluded due to lack of political data. Alaska and Hawaii joined the Union in 1959, but dont have complete data until 1960. First differencing removes one year. This leaves eight years of data. For eight states for 46 years, there should be 376 cases. However, Michigan is missing data for the number of state employees for five of those years. An additional year lost to first differencing for the first year of Michigan data leaves 370 cases. In the years 19601969, there are 49 states that should have data for 10 years. Hawaii is missing data for 1960 and 1961 for several variables, which means first differenced measures end up being missing for subsequent years, resulting in two missing cases, with an additional case missing due to perfect collinearity with year fixed effects in Alabama. Data are missing for two years in the 1970s (1973 and 1974), resulting in data for 49 states for eight years; two data points are missing due to missing data for Alabama because of perfect collinearity with year fixed effects. Imputing missing data for the time series when volatility in the time series is crucial to the analysis is an area of active development, with no clear guidelines emerging from the literature. In general, a Kalman smoother can be used in these circumstances, but will typically just replicate previous patterns (Chatfield, 2016). Given that we are looking for changes in patterns, we do not use multiple imputation for missing data.
5. Ideally, we would be able to include additional control variables in our analysis, such as those used in prior research (Delaney & Doyle, 2007, 2011, 2014). In particular, a control for postsecondary student enrollment would help specify the model. However, neither states nor the federal government consistently collected this information prior to the HEGIS data system, which later became IPEDS. We acknowledge this as a limitation, although we think that the benefits of the analysis that we conduct are outweighed by the need for additional control variables.
Betts, J. R., & McFarland, L. L. (1995). Safe port in a storm: The impact of labor market conditions on community college enrollments. The Journal of Human Resources, 30(4), 741765.
Bureau of Labor Statistics. State and area employment, hours, and earnings. Retrieved from http://data.bls.gov/PDQ/outside.jsp?survey=sm on June 21, 2016.
Chambers, M. M. (1961). Appropriations of state tax funds for operating expenses of higher education 1960-61 (Technical Report, Joint Office of Institutional Research). Washington, DC . (N.B. Over 40 editions of Grapevine were used for the data in this studythese references have been omitted for the sake of brevity.)
Chatfield, C. (2016). The analysis of time series: An introduction. CRC Press.
Clotfelter, C. (1976). Public spending for higher education: An empirical test of two hypotheses. Public Finance, 31(2), 177195.
Dar, L., & Lee, D. W. (2014). Partisanship, political polarization, and state higher education budget outcomes. The Journal of Higher Education, 85(4), 469498.
Delaney, J. A., & Doyle, W. R. (2007). The role of higher education in state budgets. In D. E. Heller & K. M. Shaw (Eds.), State postsecondary education research: New methods to inform policy and practice (pp. 5576). Sterling, VA: Stylus Publishing.
Delaney, J. A., & Doyle, W. R. (2011). State spending on higher education: Testing the balance wheel over time. Journal of Education Finance, 36(4), 343368.
Delaney, J. A., & Doyle, W. R. (2014). State spending on higher education capital outlays. Research in Higher Education, 55(5), 433466.
Ewell, P. T. (2009). Shovel-ready data: The stimulus package and state longitudinal data system. Assessment update: Progress, trends, and practices in higher education, 21(5), 1112.
Ewell, P. T., & LOrange, H. (2009). The ideal state postsecondary data system: 15 essential characteristics and required functionality. Boulder, CO: State Higher Education Executive Officers.
Ewell, P. T., Schild, P. R., & Paulson, K. (2003). Following the mobile student: Can we develop the capacity for a comprehensive database to assess student progression? (Lumina Foundation for Education Research Report). Boulder, CO: National Center for Higher Education Management Systems.
Fiorina, M. P. (1991). Divided government in the states. PS: Political Science & Politics, 24(4), 646650.
Goldin, C., & Katz, L. F. (1999). The shaping of higher education: The formative years in the United States, 1890 to 1940. Journal of Economic Perspectives, 13(1), 3762.
Greene, W. H. (2003). Econometric analysis. Upper Saddle River, NJ: Prentice Hall.
Heck, R. H., Lam, W. S., & Thomas, S. L. (2012). State political culture, higher education spending indicators and undergraduate graduation outcomes. Educational Policy, 0895904812453996.
Hovey, H. A. (1999). State spending for higher education in the next decade: The battle to sustain current support. San Jose, CA: California State Policy Research, Inc.
Humphreys, B. R. (2000). Do business cycles affect state appropriations to higher education? Southern Economic Journal, 67(2), 398413.
Kane, T. J., Orszag, P. R., & Gunter, D. L. (2003). State fiscal constraints and higher education spending: The role of Medicaid and the business cycle. Washington, DC: Brookings Institution.
Koshal, R. K., & Koshal, M. (2000). State appropriation and higher education tuition: What is the relationship? Education Economics, 8(1), 8189.
Leslie, L. L., & Ramey, G. (1986). State appropriations and enrollments: Does enrollment growth still pay? The Journal of Higher Education, 57(1), 119.
McLendon, M. K., Hearn, J. C., & Mokher, C. G. (2009). Partisans, professionals and power: The role of political factors in state higher education funding. The Journal of Higher Education, 80(6), 686713.
McLendon, M. K., Tandberg, D. A., & Hillman, N. W. (2014). Financing college opportunity factors influencing state spending on student financial aid and campus appropriations, 1990 through 2010. The ANNALS of the American Academy of Political and Social Science, 655(1), 143162.
National Bureau of Economic Research. Business cycle expansions and contractions. Retrieved from http://www.nber.org/cycles/cyclesmain.html on June 21, 2016.
Nicholson-Crotty, J., & Meier, K. J. (2003). Politics, structure, and public policy: The case of higher education. Educational Policy, 17(1), 8097.
Okunade, A. A. (2001). What factors influence state appropriations for public higher education in the United States? Journal of Education Finance, 30(2), 123138.
Palmer, J. C. (2005). Grapevine survey of state higher education tax appropriations for fiscal year 2005 (Technical Report, Center for the Study of Educational Policy). Normal, IL.
Peterson, R. G. (1976). Environmental and political determinants of state higher education appropriations policies. The Journal of Higher Education, 47(5), 523542.
Rizzo, M. J. (2003). A (less than) zero sum game? State funding for public education: How public higher education institutions have lost (Unpublished doctoral dissertation). Cornell University, Ithaca, NY.
State Higher Education Executive Officers. (2016). State higher education finance: FY 2015. Retrieved from http://www.sheeo.org/sites/default/files/project-files/SHEEO_FY15_Report_051816.pdf on June 21, 2016.
Stone, J. A. (2016). A poison Pell for public colleges? Pell Grants and funding for public colleges in the U.S. (MPRA Paper No. 71761). Retrieved from https://mpra.ub.uni-muenchen.de/71761/ on June 24, 2016.
Tandberg, D. A., & Ness, E. C. (2011). State capital expenditures for higher education: Where the real politics happens. Journal of Education Finance, 36(4), 394423.
Toutkoushian, R. K., & Hollis, P. (1998). Using panel data to examine legislative demand for higher education. Education Economics, 6(2), 141157.
United States Bureau of the Census. Statistical Abstracts of the United States. Retrieved from http://www.census.gov/compendia/statab/. (N.B. Over 55 editions of the Statistical Abstracts of the United States were used for the data in this studythese references have been omitted for the sake of brevity.)
United States Bureau of the Census. (19601988). Statistical Abstract of the United States (Technical Report). Washington DC: Author.
United States Bureau of Commerce, Bureau of Economic Analysis. (2009). Regional economic accounts data. Retieved on March 15, 2009, from http://www.bea.gov/regional/index.htm#state.
U.S. Department of Education, Institute of Education Statistics, National Center for Education Statistics. (2006). Digest of education statistics 2006. Retrieved on February 23, 2015, from https://nces.ed.gov/programs/digest/2006menu_tables.asp.