Schoolhouse Democracy: Public Opinion and Education Spending in the States

Abstract

Using new estimates of state-level public opinion, I explore the relationship between support for increased education spending and statewide per-pupil expenditures from 1986 to 2013. In the 1980s, there was a modest, positive relationship between public opinion and actual spending: States with greater support for increased education spending tended to have slightly higher per pupil expenditures. Over the next three decades, this relationship reversed. States with relatively low per-pupil expenditures tended to increase their spending at a slower rate despite steady growth in support for more spending. As a result, public opinion and education spending became inversely related. By the end of the time series, states with greater support for increased education spending tended to spend less per pupil. The changing distribution of local, state, and federal sources of education spending partially explains this pattern. As federal education expenditures rose, some states spent proportionally less from state and local sources, resulting in smaller overall spending increases in those states.

Keywords

descriptive analysis education finance educational policy politics of education public opinion survey research

A lengthy tradition in political science examines the link between public opinion and the actual policies that governments produce. A positive relationship between the two implies that places with greater public support for a set of policies are more likely to see those policies enacted. A strong opinion-policy connection is not necessarily an unalloyed good. The merits of public opinion often depend on the perspective of the observer. At different times, in different places, and on different issues, the preferences of the majority can be sophisticated or imprudent, well informed or misguided, community-minded or self-serving. However, if a one-to-one relationship between opinion and policy is sometimes undesirable, then the absence of any connection could be cause for greater concern. Political scientists differ on the necessary and sufficient conditions for democracy,¹ but the most basic prerequisite is straightforward: “Unless mass views have some place in the shaping of policy, all the talk about democracy is nonsense” (Key, 1961, p. 7).

This study explores the extent to which differences in education spending reflect differences in public opinion. Specifically, I consider the relationship between public opinion on education spending and K–12 per-pupil expenditures in the American states, the units of government with constitutional responsibility to provide public education and that supply the plurality of education spending in the United States (National Center for Education Statistics, 2015; Parker, 2016). I examine how the citizens of each state vary in their support for increased spending on public education, and I compare this to actual statewide per-pupil expenditures. I organize this investigation around three basic questions. First, in a given year, do states with greater support for education spending tend to spend more per pupil? Second, are increases in support for education spending followed by increases in per-pupil expenditures? And third, how do these relationships change over time?

I aim to provide a descriptive account of the relationship between public opinion and education spending. I seek to understand whether state residents who want more money spent on public education are, in fact, getting additional expenditures per pupil. I do not, however, examine whether changes in public opinion cause changes in education spending.

I find that both statewide public opinion in favor of greater education spending and statewide per-pupil expenditures have increased from the mid-1980s onward. From 1986 to 1991, the relationship between public opinion and education spending was positive: States with greater support for more education spending tended to spend more per pupil. Similarly, within states over time, higher levels of support for education spending were typically met with higher expenditures 2 years later. However, these relationships have shifted over time. Initially lower spending states increased per-pupil expenditures at a slower rate than their initially higher spending counterparts despite concurrent increases in support for additional spending. As a result, the opinion-spending relationship gradually became negative: Greater support for more education spending became associated with relatively smaller per-pupil expenditures. The changing distribution of local, state, and federal sources of education spending partially explains this pattern. As federal education expenditures rose, some states spent proportionately less from state and local sources, resulting in smaller overall spending increases in those states.

These results not only contribute to the academic literatures on education policy and finance, they also potentially shed light on contemporary political battles over spending on public education. For example, in 2018 and the beginning of 2019, major teacher strikes erupted in Arizona, California, Colorado, North Carolina, Oklahoma, and West Virginia (Fernandez Campbell, 2019). In all six cases, teachers and their allies sought higher pay and additional resources for public schools. Notably, these strikes all occurred in states where average per-pupil expenditures (of which teacher salaries and benefits make up the largest single expense) increased relatively slowly since the 1980s while public opinion in favor of increased education spending rose more rapidly. Although this study does not demonstrate that the growing disjuncture between public opinion and education spending directly motivated the teacher strikes, it provides a possible explanation for why the political conditions were favorable for such action. As education leaders, policymakers, advocates, and the public seek to understand and navigate the current debates over school spending, they should consider how public opinion might facilitate or inhibit various political strategies.

Literature Review

This study follows from the normative expectation that public policy, including education policy, ought to have at least a passing relationship with public opinion. In the political science literature, this relationship is referred to as policy congruence or policy responsiveness (Beyer & Hänni, 2018). Congruence refers to the extent to which policy is consistent with public opinion at a single point in time. Congruence occurs when, at a given point in time, places with more (or less) support for a set of policies are more (or less) likely to have those policies enacted. Responsiveness refers to the extent to which governments respond to changes in public opinion over time by shifting policy in the same direction. Responsiveness occurs when, in a given place over time, increasing (or decreasing) support for a set of policies is associated with an increasing (or decreasing) likelihood of those policies being enacted. Both congruence and responsiveness fall under the broader concept of policy representativeness.

The existing evidence in favor of policy representativeness is robust. At the national level, Page and Shapiro (1983) investigated the extent to which changes in federal spending preferences on a variety of issues are succeeded by actual changes in federal expenditures and found corresponding movement roughly two-thirds of the time. Erikson, MacKuen, and Stimson (2002) shifted away from the study of individual policies and spending levels, and suggested that it might be more useful to analyze the relationship between overall political mood (a population-level ideology measure) and an ideologically indexed composite of issues. Given Americans’ generally low levels of policy knowledge, they believed it made more sense to look at larger ideological responsiveness. They found that an increase in liberal mood is typically followed by more liberal laws and more public spending and an increase in conservative mood is typically followed by more conservative laws and less public spending.

At the state level, Erikson, Wright, and McIver (1993) demonstrated that state policies are also consistent with the ideological preferences of their citizens. States with more liberal residents tend to have more liberal policies, and states with more conservative residents tend to have more conservative policies. Lax and Phillips (2009a, 2012) estimated the percentage of adults in each state that support a range of policies at a single point in time. Across multiple issue areas, they found evidence that greater support for a policy is associated with a greater likelihood of that policy being enacted, but they also found a “democratic deficit”: Actual policy matches majority opinion only about half of the time.

Tausonovitch and Warshaw (2014) revealed that the level of liberalism/conservatism in local public opinion corresponds with the ideological orientation of cities’ policies across a range of issues. Einstein and Kogan (2016) found that as cities become more Democratic (i.e., a higher Democratic vote share in presidential elections), they also tend to increase public spending across many service areas.

Two studies explicitly examined education spending representativeness. Berkman and Plutzer (2005) made use of the roughly 10,000 elementary and secondary school systems in the country to construct an analysis of congruence at one of the smallest and most familiar levels of government: local school districts. On average, higher rates of districtwide support for increased education spending are associated with higher local per pupil expenditures. Pacheco (2013) explored responsiveness in state education (including higher education, adult education, and aid to private schools) and welfare spending. She demonstrated that increases in support for education and welfare spending are typically followed by increases in actual spending.

I contribute to this literature by conducting the first state-level policy representativeness analysis with a specific focus on K–12 public education spending. I extend Berkman and Plutzer’s (2005) work on congruence to the state level. District spending levels, although increasingly shaped by state and federal regulations, still depend to some degree on local property values, the outcomes of school board elections, and the results of local school funding levies. There are fewer opportunities to influence state education spending levels at the ballot box, and although households may choose to live in certain school districts based on education resources, residential sorting is less likely to occur across state lines (Tiebout, 1956). It remains an open question whether there is evidence of education spending congruence at the state level. I also extend Pacheco’s (2013) work on state-level responsiveness by examining the special case of K–12 spending. Lastly, I consider how both congruence and responsiveness have shifted over time.

Methodology

Analytic Approach

The primary obstacle in the study of state-level education spending representativeness is the absence of public opinion surveys featuring a relevant and consistently worded question in each state and year. I rely on a relatively novel approach to the estimation of state-level public opinion: multilevel regression and poststratification or MRP (Kastellec, Lax, & Phillips, 2016; Lax & Phillips, 2009b; Pacheco, 2011; Park, Gelman, & Bafumi, 2004). MRP contains three steps: (a) modeling education spending preferences as a function of demographic and geographic predictors with nationally representative survey data, (b) using the coefficients from this model to generate predicted probabilities of supporting increased education spending for a range of demographic-geographic categories, and (c) weighting these predicted probabilities by U.S. Census counts of these categories in each state and year. This procedure produces 50 estimated levels of statewide support for increased education spending for each year from 1984 to 2013. With these data in hand, the fundamental question posed by this study—Is state education spending reflective of the preferences of the public?—becomes empirically tractable.

The first step begins with an analysis of the demographic and geographic predictors of support for increased education spending. I fit a multilevel logistic regression equation in which support for increased education spending (dichotomous) is a function of education attainment (less than high school, high school, some college, or college+), race (White or person of color), sex (female or male), age (18–44, 45–64, or 65+), U.S. state of residence, the percentage of state residents who identify as liberal, and U.S. Census region (Midwest, Northeast, South, or West), with varying intercepts by state. Lax and Phillips (2009b) established that a simple set of individual demographic characteristics along with one or two state or regional characteristics are sufficient for generating state-level MRP estimates when those variables are important predictors of the political attitude in question. Berkman and Plutzer (2005) demonstrated that race, gender, age, and education are all predictive of support for increased education spending.

I conducted a separate multilevel model for each year in the time series, pooling data from years $t \pm 4$ to increase the size of the state subsamples (e.g., for the 1994 model, I pooled data from 1990–1998). Next, using the coefficients from these models, I generated predicted probabilities of support for increased education spending for 48 different demographic categories (each education attainment, race, sex, and age combination) in each state in each year for a total of 72,000 separate categories. To calculate these predicted probabilities, I employed the following equation:

\Pr (Y = 1) = l o g i t^{- 1} (\begin{array}{l} E d_{i s r} + R a c e_{i s r} + S e x_{i s r} + A g e_{i s r} \\ + S t a t e_{s r} + L i b_{s r} + R e g i o n_{r} \end{array}),

(1)

in which each variable represents the applicable logistic regression coefficient for individual type i in state s and region r.

Finally, I poststratified by weighting these values using data from the U.S. Census Bureau. I multiplied the predicted probabilities by counts of each demographic category in each state and year to generate estimates of the total number of supporters. I then divided by the total number of adults in each state and year to generate estimates of support for increased education spending.

To establish the face validity of these values, I compared them to the national trend line of support for increased education spending (which does not rely on MRP). For a more formal assessment of the validity of the MRP estimates, I compared them to the results of an older, well established, but less precise method for estimating state-level public opinion: survey aggregation. Both approaches produced similar estimates of support for increased education spending (see Appendix for more detail).

To evaluate congruence, I employed the following ordinary least squares regression equation:

Y_{s t} = β_{0} + β_{1} S u p p o r t_{s t} + Y e a r_{t} + ϵ_{s t},

(2)

where Y is the outcome (state per-pupil expenditures), Support is the percentage of the state population that supports increased education spending, Year are year fixed effects, and ϵ is the error term in state s and year t.

To evaluate responsiveness, I modified the equation slightly:

Y_{s t} = β_{0} + β_{1} S u p p o r t_{s t} + S t a t e_{s} + ϵ_{s t},

(3)

where Y is the outcome (state per-pupil expenditures in year t + 2 or the change in per-pupil expenditures from year t to year t + 2), Support is the percentage of the state population that supports increased education spending in year t, State are state fixed effects, and ϵ is the error term in state s and year t.²

To document how congruence and responsiveness have changed over time, I reconducted the aforementioned analyses using data from six separate 5-year periods: 1984–1988, 1989–1993, 1994–1998, 1999–2003, 2004–2008, and 2009–2013.

I also examined three state-level predictors of changes in education spending over time with the following model:

Y_{s} = β_{0} + β_{1} I n i t i a l P P E_{s} + β_{2} C h a n g e S u p p o r t_{s} + β_{3} C h a n g e F e d_{s} + ϵ_{s},

(4)

where Y is the outcome (the change in per-pupil expenditures from 1986 to 2013), InitialPPE is per-pupil expenditures in 1986, ChangeSupport is the change in the percentage of the state population that supports increased education spending from 1986 to 2013, ChangeFed is the change in the proportion of education spending from federal sources from 1997 to 2013, and ϵ is the error term in state s.

Data

I relied on the General Social Survey (GSS) for data on public opinion regarding education spending. MRP-generated estimates of state-level public opinion over an extended time series require longitudinal survey data that contain individual demographic variables, state identifiers, and a consistently worded question on attitudes toward education spending. The GSS meets all of these basic criteria.

However, the GSS also contains features that generate three central challenges for this study. The first concern relates to question wording. Minor changes in the structure of survey questions—even if the content is logically equivalent—can have nontrivial effects on the distribution of responses (Schuman & Presser, 1981). The full GSS question reads: “Are we spending too much, too little, or about the right amount on improving the nation’s education system? (Too Little, About Right, Too Much)” This question inverts the conventional answer options, asking respondents if we spend too little as opposed to asking them if spending should be increased. The GSS also asks respondents about spending on the nation’s education system rather than spending in their state. Berkman and Plutzer (2005) demonstrated that the answers to the GSS question largely mimic the answers to more context-specific questions posed by smaller, nonlongitudinal surveys. Although absolute levels of support for education spending differ from survey to survey, the relative degree of support across racial, education, and age categories is similar regardless of whether the survey emphasizes “your community’s public schools,” “state and local spending,” or “the nation’s education system.” Berkman and Plutzer argued that this consistency indicates that the GSS question taps into a broader notion: a general taste for education spending.

The second concern about the GSS relates to its sampling design. The GSS employs cluster random sampling to obtain a nationally representative sample while minimizing some of the logistical challenges of reaching respondents where they live (the GSS conducts in-person interviews). As a result, the respondents selected within each state are not necessarily a representative sample of that state. I addressed this issue in four ways. First, to estimate state-level public opinion in year $t$ t, I employed a simple moving average that pools survey data from multiple years ( $t \pm 4$ ) (see Pacheco, 2011). This approach both increases my subsamples in each state and decreases the likelihood that all respondents from the same state are drawn from the same cluster. Second, when modeling support for increased education spending, I used a multilevel approach that relies on data from the whole sample while allowing the intercept to shift slightly for each state. This attenuates the influence of an unusual group of respondents from a single state in favor of patterns observed across all respondents. Third, my models included a state-level predictor (the percentage of self-identified liberals in a state) and a set of U.S. Census region indicators, both of which shape the magnitude and direction of the relationships between state of residence and support for increased education spending. Fourth, the state-level estimates of public opinion are also influenced by the demographic compositions of those states, which come into play during the poststratification stage of MRP and which are independent of survey samples. Collectively, these safeguards make it less likely that the final state-level public opinion estimates are skewed by an unusual subsample in a given state and year (see Stollwerk, 2013, for more information on using MRP with cluster-sampled polls).

The final concern relates to the question’s omission of any policy tradeoffs. Increases in education spending would typically come at the expense of less spending on other policy domains, higher taxes, or increases in government deficits. Indeed, support for increased education spending declines when these tradeoffs are mentioned in the question itself (Schueler & West, 2016). The estimates of public opinion captured by the GSS question may be somewhat inflated by this omission. On the other hand, Berkman and Plutzer (2005) found that the relative support for increased education spending between states is fairly consistent across question wordings. In other words, my estimates of public opinion in California and Kansas using the GSS question may be higher than analogous estimates using a question that identifies policy tradeoffs, but I would expect California to have a higher level of support than Kansas in both versions. Because my analyses rely primarily on between-state and within-state comparisons, the absolute levels of support for increased education spending are less pertinent.

Because of the GSS’s use of identical question wording over multiple survey administrations, I could generate both national and statewide estimates of support for increased education spending for each year from 1984 to 2013. GSS data are available for every even-numbered year and some odd-numbered years before 1994. GSS data are also available over a longer time series, but at the time of analysis, contemporaneous statewide per-pupil expenditure data were only available from 1986 to 2013. Subjects who indicated that we are spending too little on education were coded as 1. All other responses were coded as 0. About 5% of subjects did not respond to the education spending question. I dropped these observations from the analysis.

For the measures of education spending, I relied on the National Center for Education Statistics Common Core of Data, which includes national and statewide total current per-pupil expenditures for K–12 public schools from the 1986–1987 academic year to the 2013–2014 academic year. The Common Core of Data also includes data on the proportion of state education spending from federal sources from 1997 onward. I adjusted all per-pupil expenditure values for inflation (2017 dollars) and for regional differences in cost of living using the U.S. Bureau of Labor Statistics Consumer Price Index. These values include all expenditures for public elementary and secondary education from local, state, and federal sources (excluding spending on non–public education, school construction, debt financing, and community services).

I did not restrict this analysis to spending from state sources alone. States vary in the percentages of education spending that come from local, state, and federal sources, and state spending decisions potentially account for these differences (Berkman & Plutzer, 2005). When descriptively evaluating the relationship between public opinion and education spending, it is necessary to include spending from all levels of government. However, the following analysis also examines how the changing distributions of local, state, and federal sources of education spending relate to overall changes in statewide per-pupil expenditures.

For a measure of the percentage of adults in each state that identifies as liberal, I used data generated by Pacheco (2011) for her analysis on state public opinion over time.

To assemble U.S. Census counts for each demographic-geographic category, I used the U.S. Census Integrated Public Use Microdata Series (IPUMS). For 1984–1994 counts, I relied on the 1990 Census (1% sample); for 1995–2004 counts, I relied on the 2000 Census (1% sample); and for 2005–2013 counts, I relied on the 2010 American Community Survey (5% sample).

Findings

Estimating State-Level Public Opinion

The relationships between a series of demographic characteristics and support for increased education spending are displayed in Table 1. Support for increased education spending is positively associated with increases in education attainment, identifying as a racial group other than White, identifying as female, and living in a state with a higher percentage of self-identified liberals. Older individuals tend to express less support for increased education spending than their younger counterparts. All else equal, individuals in the South generally have greater support for increased education spending than those in other U.S. Census regions. These relationships are consistent with those identified by Berkman and Plutzer (2005).³

Table 1

Predictors of Public Opinion on Education Spending

	Support for Increased Education Spending
High school	0.23*** (0.03)
Some college	0.44*** (0.03)
College+	0.48*** (0.03)
Person of color	0.30*** (0.03)
Female	0.25*** (0.02)
Age 45–64	−0.32*** (0.02)
Age 65+	−0.78*** (0.03)
State % liberal (standardized)	0.05 †(0.03)
Northeast	−0.05 (0.08)
South	0.20*** (0.06)
West	0.08 (0.07)
Year fixed effects	Yes
Varying intercepts by state	Yes
Akaike Information Criterion	53,289
Bayesian Information Criterion	53,595
N	45,912

Note. Values are multilevel logistic regression coefficients with standard errors in parentheses; units are individual survey respondents; education attainment categories compared to less than high school; age categories compared to age 18–44; U.S. Census regions compared to Midwest.

†

p < .10. ***p < .001.

I fit an analogous model for each year from 1984 to 2013, pooling available survey data from years $t \pm 4$ . Using MRP, I employed the coefficients from these models in conjunction with U.S. Census counts of each demographic-geographic category to generate estimates of statewide support for increased education spending for each state and year. These estimates are displayed in Figure 1. Statewide support for increased education spending ranged from about 54% in 1984 to about 81% in 2009. Since 1984, a majority of adults in all states supported increased education spending, but there was considerable variation in the level of support between states. Support for increased education spending was highest in liberal and racially diverse states such as California, Hawaii, and Maryland. Southern states with large communities of color such as Alabama, Florida, and Mississippi also tended to have relatively high support for increased education spending. Support was lowest in states with older, Whiter, and more conservative populations, such as Kansas, Nebraska, and the Dakotas. Relatively low support for increased education spending also appeared in politically liberal but older and Whiter states such as Maine, Vermont, and Wisconsin. Overall, state-level public opinion tracked the national trend (seen in bold in Figure 1), with a consistent increase until 2004 and relatively steady support thereafter. The distribution of state-level public opinion narrowed modestly over time. There was slightly more variation in public opinion in the earlier half of the time series. Support for increased education spending rose faster in states near the bottom of the distribution than in states with higher initial support.

Figure 2 displays the state per-pupil expenditure trends from 1986 to 2013 as well as the average per-pupil expenditures across all states in each year. Education spending tended to be higher in politically liberal states, but there were some notable exceptions. Per-pupil expenditures were also relatively high in sparsely populated Alaska and Wyoming. Alternatively, education spending was lowest in Southern and Western states such as Arizona, Idaho, Mississippi, Tennessee, and Utah. The average education spending trend followed the same trajectory as the national public opinion trend. It rose more or less continuously from 1986 to 2009. The next 3 years witnessed a relative decline in education spending, followed by a slight rebound. In contrast to public opinion, the spending distribution widened over time. There was more variation in state per-pupil expenditures in the latter half of the time series. States with relatively low spending in the mid-1980s tended to increase spending at a slower rate than their counterparts with higher initial expenditures.

Representativeness in State Education Spending

Figure 3 displays the relationships between support for increased education spending and state per-pupil expenditures within each year (congruence). The linkage gradually changed over time. In the 1980s, there was a modest, positive relationship between public opinion and actual spending levels: States with greater support for increased education spending tended to spend more per pupil. This pattern reversed by the mid-1990s, growing increasingly negative through the 2000s. For much of the latter half of the time series, states with greater support for more education spending tended to spend less per pupil. However, there were many states that deviated from the overall pattern. In all years, there were states with relatively high levels of support for increased education spending but relatively low spending levels. The reverse was also true: Many states with relatively low levels of support spent generously by comparison.

The relationship between public opinion and education spending within states over time (responsiveness) was more consistent. Greater support for increased education spending was associated with higher state per-pupil expenditures 2 years later. Figure 4 reveals a visually apparent positive relationship between the two in most states. However, there is reason to be skeptical about whether this descriptive evidence for responsiveness represents a causal relationship (i.e., that states were deliberately responding to changes in public opinion with changes in per-pupil expenditures). Rather, it is plausible that both state-level public opinion and spending levels were reacting to the same external pressures.

I explore these relationships in more detail in Table 2. The inclusion of year fixed effects allows Model 1 to quantify the overall relationship displayed in Figure 3. On average, there was a negative relationship between public opinion and education spending between states in a given year: A 1 percentage point increase in support for more education spending was associated with a $98 decrease in per-pupil expenditures. The inclusion of state fixed effects allows Model 2 to quantify the overall relationship displayed in Figure 4. On average, there was a positive relationship between public opinion and education spending within states over time: A 1 percentage point increase in support for more education spending in a given state was associated with a $298 increase in per-pupil expenditures 2 years later. The results in Models 1 and 2 reiterate the graphical evidence against congruence and in favor of responsiveness. Although this may seem contradictory, these patterns can coexist. The remainder of the analysis will show that although public opinion in favor of additional education spending and actual per-pupil expenditures have both increased over time (generating the appearance of responsiveness within states), the growth in support for additional spending and the growth in per-pupil expenditures have both varied by state (creating the conditions for public opinion to diverge from spending levels between states at a given point in time).

Table 2

Representativeness in State Education Spending

	Congruence	Responsiveness
	State PPE	State PPE 2 Years Later	Δ State PPE 2 Years Later
	(1)	(2)	(3)
% Support	−97.81*** (19.89)	298.47*** (11.12)	−5.63 (4.81)
Year fixed effects	Yes	No	No
State fixed effects	No	Yes	Yes
R ²	.28	.80	.06
$N$ N	1,400	1,300	1,300

Note. Values are ordinary least squares regression coefficients with standard errors in parentheses; units are state-years; PPE = per-pupil expenditures.

***

p < .001.

In Model 3, I examined whether increases in support for more education spending were followed by larger increases in state per-pupil expenditures. Within states, there was no relationship between public opinion and the rate at which expenditures increased over the next 2 years. This finding complicates the pattern of evidence in favor of responsiveness. If there were a simple causal relationship between public opinion and education spending within states, then we might expect to see greater demand for increased education spending followed by larger spending increases. The absence of a relationship in this context strengthens the case for the argument that although both public opinion and education spending were rising, the former did not necessarily cause the latter.

Table 3 disaggregates these relationships by 5-year periods. Panel A documents the reversal of congruence over time. Whereas public opinion and education spending were positively related in 1984–1988, the relationship turned increasingly negative over the next 25 years. Panel B explores the changing link between public opinion and education spending within states. The evidence in favor of responsiveness was strongest in 1994–1998. There was no within-state opinion-policy relationship in the last 10 years of the time series.

Table 3

Representativeness in State Education Spending Over Time

Panel A. Congruence	State PPE
Panel A. Congruence	(1984–1988)	(1989–1993)	(1994–1998)	(1999–2003)	(2004–2008)	(2009–2013)
% Support	108.00* (48.17)	43.38 (40.74)	−74.54* (33.82)	−177.42*** (46.64)	−238.63*** (61.62)	−292.42*** (58.13)
Year fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
State fixed effects	No	No	No	No	No	No
R ²	.05	.01	.04	.08	.07	.10
N	150	250	250	250	250	250
Panel B. Responsiveness	State PPE 2 Years Later
Panel B. Responsiveness	(1984–1988)	(1989–1993)	(1994–1998)	(1999–2003)	(2004–2008)	(2009–2013)
% Support	116.28*** (26.06)	51.14*** (10.84)	283.18*** (33.65)	132.46*** (27.00)	−8.91 (25.93)	22.70 (18.16)
Year fixed effects	No	No	No	No	No	No
State fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
R ²	.99	.99	.95	.97	.98	.99
N	150	250	250	250	250	150

Note. Values are ordinary least squares regression coefficients with standard errors in parentheses; units are state-years; PPE = per-pupil expenditures.

p < .05. ***p < .001.

What has been driving this decline in representativeness? Figure 5 considers three state-level predictors of the change in per pupil expenditures from 1986 to 2013: initial spending in 1986, the change in public opinion over time, and the change in the proportion of education spending from federal sources over time (these relationships are also quantified in Table 4). The first plot displays the different paths taken by initially high-spending and initially low-spending states. States with higher per-pupil expenditures in 1986 saw larger total increases in expenditures over the time series. The second plot indicates that states with larger increases in support for additional spending tended to experience smaller overall increases in expenditures (note that this relationship narrowly exceeds the .10 threshold for statistical significance). Both patterns persist after adjusting for the other predictor. In short, low-spending states in 1986 tended to expand their education budgets at a slower rate despite increases in support for more spending. This dynamic resulted in the reversal of the opinion-spending relationship over time.

Table 4

Predictors of Education Spending Increases

	Δ State PPE (1986–2013)
	(1)	(2)	(3)	(4)	(5)
State PPE in $1000s (1986)	273.15* (116.50)			268.61*(114.87)	89.03(114.84)
Δ % Support (1986–2013)		−153.42(99.50)		−147.75(95.20)	−127.31(85.54)
Δ % Federal (1997–2013)			−614.16*** (139.47)		−546.31*** (154.66)
R ²	.10	.05	.29	.15	.33
N	50	50	50	50	50

Note. Values are ordinary least sqaures regression coefficients with standard errors in parentheses; units are states; PPE = per-pupil expenditures.

*p < .05. ***p < .001.

The changing distribution of local, state, and federal sources for education spending within states partially explains this pattern. Federal spending on K–12 education has increased over time, but federal contributions as well as local responses to those contributions have varied by state (National Center for Education Statistics, 2015). The third plot illustrates that states with larger increases in the proportion of education spending from federal sources experienced smaller overall increases in per-pupil expenditures, suggesting that state and local spending did not rise commensurately with federal spending over the same period. Moreover, this relationship persists among those states with similar initial spending levels and similar changes in public opinion over time. This dynamic is consistent with the overall finding that public opinion and overall education spending diverged despite increases in both over time.

At first glance, this pattern seems to run afoul of the “Supplement-not-Supplant” provision of Title 1 of the Every Student Succeeds Act, which ensures that federal funds for low-income schools are additive and do not take the place of state and local funds for the same purposes (U.S. Department of Education, 2016). However, it is important to keep in mind that education spending from all sources has generally risen across the board. The relative increase in the proportion of federal funds does not necessarily suggest that states reduced their own expenditures on low-income schools (although it does not rule this possibility out either). Rather, this relationship simply indicates that in some locations, increases in state and local spending did not keep pace with increases in federal spending.

It is worth noting that the six states that experienced major teacher strikes in 2018 and the beginning of 2019—Arizona, California, Colorado, North Carolina, Oklahoma, and West Virginia—appear in similar locations throughout Figure 5. All six states had comparatively small increases in education spending since the 1980s. They also tended to have lower initial per-pupil expenditures, moderate increases in support for more spending, and larger increases in the proportion of education spending from federal sources (indicating that state and local spending grew relatively slowly). It may be the case that the divergence between public opinion and per-pupil expenditures in these states facilitated the emergence of political movements demanding more spending on public education. Although this study does not directly test the proposition that these factors contributed to the strikes, it helps to explain why the political conditions were favorable to such action.

Conclusions

When asked about public education expenditures in the United States—without considering tradeoffs such as higher taxes or reduced services in other areas—a majority of Americans are in favor of spending more on public schools. This has been the case since at least 1984, and support has generally grown over time. Per-pupil expenditures have also increased during this period. The national trend lines of public opinion and education spending essentially overlap: They have risen in tandem from the mid-1980s onward.

At the state level, this relationship is more complex. At the start of the time series presented here, states with greater support for increased education spending tended to spend more per pupil. Moreover, as support for increased education spending rose, so did education spending levels a few years later. These findings are consistent with a large body of research exploring policy congruence and responsiveness across many issue domains. However, over the next three decades, the states with initially higher per-pupil expenditures tended to increase their spending at a faster rate than the states with initially lower per-pupil expenditures. Concurrently, public opinion in favor of more education spending rose across the board. As a result, public opinion and education spending became inversely related: States with greater support for increased education spending tended to spend less per pupil.

One potential explanation for this pattern focuses on the varying role that federal spending has come to play in states’ K–12 education budgets. Federal spending on education has increased over time, and federal dollars now constitute a slightly larger proportion of overall education spending. However, the growth in the proportion of education spending from federal sources varied meaningfully by state. Some states increased state and local expenditures commensurately with the rise in federal expenditures, whereas others, for a host of different reasons, did not. Consistent with this pattern, the variation in state-level per-pupil expenditures has widened since the 1980s. In many states, support for additional education spending appears to have outpaced the growth in actual expenditures.

It is important to be cautious about applying strong causal interpretations to these relationships. The appearance of representativeness—or the lack thereof—may also result from the reactions of both public opinion and education spending to the same set of external pressures (underlying economic conditions, demographic changes, etc.). If one is interested in a straightforward “report card” on the condition of state democracy with respect to education spending, then the concerns about causal inference are of secondary importance. The central question—as it is in this article—is simply whether education spending reflects public opinion. However, if one is interested in improving policy representativeness, then understanding the causal linkages becomes paramount. Future work in both education research and political science should focus on identifying exogenous shifts in public opinion that can support causal inferences as well as the potential mechanisms through which opinion could shape policy.

There are other important limitations of this analysis. Chiefly, I rely on a relatively novel statistical approach to estimate public opinion at the state level. I establish the reasonable validity of these estimates in the Appendix, but they remain estimates nonetheless. The reader should not mistake them for unimpeachably precise measures of support for increased education spending in each state and year. In any given state, fielding a high-quality poll would provide a more definitive estimate of public opinion on this issue. This study also does not examine the many constraints that state policymakers, organized interests, and other political actors encounter when engaging in debates over resource allocation. Moreover, I only explore one element of education policy at one level of government: average statewide K–12 per-pupil expenditures. The presence or absence of a relationship between public opinion and education spending does not necessarily imply that other education policies correspond similarly to differences in public opinion. Future research should examine representativeness across a variety of other education issues.

My findings contribute to a rich literature on policy representativeness, including a small body of research on education spending. Berkman and Plutzer (2005) offered evidence of education spending congruence at the school district level. I found that the same initially holds true at the state level but the relationship shifts and ultimately reverses over time. Like Pacheco (2013), I found descriptive evidence of responsiveness in state education spending, although this relationship also appears to break down by the end of the time series. By exploring changes in these linkages over roughly three decades, I provide a more complete account of the strengths and weaknesses of education spending representativeness in the states.

The implications of this analysis for education leaders and the public are significant. Advocates of additional per-pupil expenditures—particularly in states with relatively high support and relatively low spending—could strengthen their arguments by drawing attention to the robust public backing for their position. However, in states where public opinion and per-pupil expenditures have tracked each other closely over the past three decades, it may be more challenging to rally popular support for a steeper upward spending trajectory. Alternatively, policymakers and political actors who consider current spending levels adequate or too high or who think that there are more pressing nonmonetary education reforms that ought to take priority may find more traction in states where relatively high expenditures are paired with relatively low support for additional spending. For all who seek a better understanding of the link between the public and the public schools, this analysis hopefully provides a more comprehensive picture of how Americans’ views on education spending have varied over time and place as well as the extent to which actual expenditures have reflected those preferences.

Footnotes

Appendix

To assess the validity of the multilevel regression and poststratification (MRP) estimates, I compared them to the results of an older, well-established, but less precise method for estimating state-level public opinion: survey aggregation (e.g., Erikson, Wright, & McIver, 1993). The survey aggregation method pools data from many years (10+) to generate sufficient state-level subsamples. Survey aggregation is suboptimal for studying policy representativeness because it pools data over a longer period of time, making it difficult to measure changes in opinion. Also, because it does not explicitly model policy preferences, survey aggregation is unable to minimize the complications posed by the GSS’s cluster random sampling design. However, the differences between the two approaches should be relatively modest, and they should be limited to states with smaller populations and more dramatic changes in opinion over time.

To generate survey aggregation estimates for year t, I pooled all available GSS data from years $t \pm 5$ (e.g., for 1990 estimates, I pooled all data from 1985 to 1995). Using these aggregated data, I calculated the percentage of respondents who support increased education spending in each state. I then compared these values with the corresponding MRP estimates for the same year. Figure A1 displays the results for 1990, 2000, and 2010. In all three cases, there is a strong and visually apparent association between the two sets of estimates. The states that diverge slightly from the linear trend are, as expected, the less populated states where survey aggregation ought to underperform. Moreover, the MRP estimates for these states tend to converge toward the mean, suggesting that the MRP estimates are less likely to be influenced by unusual state subsamples in a given year. Ideally, the relationship between MRP and survey aggregation estimates should produce a linear trend that falls neatly along the diagonal. All three comparisons come close to this standard.

Notes

Author

David M. Houston, PhD, is a postdoctoral research fellow at Harvard University, Kennedy School of Government, 79 John F. Kennedy Street, Cambridge, MA 02138; david_houston@hks.harvard.edu . His research interests include public opinion on education issues, racial/ethnic and economic integration, residential and school choice, and civic education.

References

Berkman

M. B.

Plutzer

(2005). Ten thousand democracies: Politics and public opinion in America’s school districts. Washington, DC: Georgetown University Press.

Beyer

Hänni

(2018). Two sides of the same coin? Congruence and responsiveness as representative democracy’s currencies. Policy Studies Journal, 46(1), 13–39.

Dahl

R. A.

(1989). Democracy and its critics. New Haven, CT: Yale University Press.

Einstein

K. L.

Kogan

(2016). Pushing the city limits: Policy responsiveness in municipal government. Urban Affairs Review, 52(1), 3–32.

Erikson

R. S.

MacKuen

M. B.

Stimson

J. A.

(2002). The macro polity. New York, NY: Cambridge University Press.

Erikson

R. S.

Wright

G. C.

McIver

J. P.

(1993). Statehouse democracy: Public opinion and policy in the American states. New York, NY: Cambridge University Press.

Fernandez

Campbell, Alexia

. (2019, February 13). A record number of US workers went on strike in 2018: Working-class Americans haven’t been this fed up with their employers since the 1980s. Vox. Retrieved from https://www.vox.com/policy-and-politics/2019/2/13/18223211/worker-teacher-strikes-2018-record

Kastellec

J. P.

Lax

J. R.

Phillips

J. H.

(2016). Estimating state public opinion with multi-level regression and poststratification using R. Retrieved from http://www.princeton.edu/~jkastell/MRP_primer/mrp_primer.pdf

Key

V. O.

Jr. (1961). Public opinion and American democracy. New York, NY: Alfred A. Knopf.

10.

Lax

J. R.

Phillips

J. H.

(2009a). Gay rights in the states: Public opinion and policy responsiveness. American Political Science Review, 103(3), 367–386.

11.

Lax

J. R.

Phillips

J. H.

(2009b). How should we estimate public opinion in the states? American Journal of Political Science, 53(1), 107–121.

12.

Lax

J. R.

Phillips

J. H.

(2012). The democratic deficit in the states. American Journal of Political Science, 56(1), 148–166.

13.

National Center for Education Statistics. (2015). Table 235.10. Revenues for public elementary and secondary schools, by source of funds: Selected years, 1919-20 through 2012-13. Retrieved from https://nces.ed.gov/programs/digest/d15/tables/dt15_235.10.asp

14.

Pacheco

(2011). Using national surveys to measure dynamic US state public opinion: A guideline for scholars and an application. State Politics and Policy Quarterly, 11(4), 415–439.

15.

Pacheco

(2013). The thermostatic model of responsiveness in the American states. State Politics and Policy Quarterly, 13(3), 306–332.

16.

Page

B. I.

Shapiro

R. Y.

(1983). Effects of public opinion on policy. American Political Science Review, 77(1), 175–190.

17.

Park

D. K.

Gelman

Bafumi

(2004). Bayesian multilevel estimation with poststratification: State-level estimates from national polls. Political Analysis, 12(4), 375–385.

18.

Parker

(2016). 50-state review: Constitutional obligations for public education. Education Commission of the States. Retrieved from https://www.ecs.org/wp-content/uploads/2016-Constitutional-obligations-for-public-education-1.pdf

19.

Schueler

B. E.

West

M. R.

(2016). Sticker shock: How information affects citizen support for public school funding. Public Opinion Quarterly, 80(1), 90–113.

20.

Schuman

Presser

(1981). Questions and answers in attitude surveys: Experiments on question form, wording, and context. New York, NY: Academic Press.

21.

Stollwerk

(2013). Estimating subnational opinion with cluster-sampled polls: Challenges and suggestions. Paper presented at the Annual Meeting of the American Political Science Association, Chicago, IL.

22.

Tausanovitch

Warshaw

(2014). Representation in municipal government. American Political Science Review, 108(3), 605–641.

23.

Tiebout

C. M.

(1956). A pure theory of local expenditures. Journal of Political Economy, 64(5), 416–424.

24.

US Department of Education. (2016). Fact sheet: Supplement-not-supplant under Title I of the Every Student Succeeds Act. Retrieved from https://www.ed.gov/news/press-releases/fact-sheet-supplement-not-supplant-under-title-i-every-student-succeeds-act