publisher colophon
abstract

We examine the correlates of district spending and revenue losses following the onset of the Great Recession and the role of fiscal federalism in mitigating these losses. We estimate whether spending and revenue declines were driven primarily by local labor market conditions or the degree of state fiscal centralization. Utilizing population level data for all public-school districts in the continental United States and a difference-in-differences strategy that models pre-recession resource trends, we find that local labor market conditions explain district spending loss in the wake of the Great Recession; in contrast, the degree of centralization in a state's education finance system is uncorrelated with declines in total district spending. Resource poor districts located in states with greater state fiscal centralization were ill-equipped to offset district spending loss, and federal fiscal stimulus did little to mitigate—and, in some cases, exacerbated—differential declines in spending resulting from local labor market shocks. These findings highlight the potentially unintended role that fiscal federalism might play in widening district spending inequality in the wake of recessionary events.

introduction

December 2007 marked the onset of an 18-month economic recession, referred to as the Great Recession, that had severe and wide-ranging economic and fiscal consequences. Unemployment increased by 5 percentage points, reaching 10% by October 2009, and these economic shocks varied widely across counties and commuting zones (Evans, et al., 2017; Hurd & Rohwedder, 2010; Wolff et al., 2011; Yagan, 2017). The fiscal effect of the Great Recession on school districts was similarly pronounced, imposing constraints on state and local funding for schools, resulting in a 3.7% decline in public school employment (Business Cycle Dating Committee, 2010; Chakrabarti & Livingston, 2013; Evans, et al., 2017; Leachman & Mai, 2014). [End Page 123]

Recent evidence documents significant cross-state variation in school spending declines following the onset of the Great Recession (Evans et al., 2017; Jackson, Wigger and Xiong, 2018). Using cross-sectional data, Evans et al. (2017) find that districts located in states which contribute a larger share of total, within-state education revenues were found to be more vulnerable to fiscal fallout (i.e., declines in education spending) in the wake of the Great Recession. Consequently, districts located in states that had taken on greater fiscal responsibility for funding K-12 public education were more vulnerable to fiscal fallout compared to districts in states in which local school districts bore greater fiscal responsibility for funding K-12 public education.

This finding has important implications for the role that state governments play in supporting educational spending at local levels. Specifically, if true, state governments would face a difficult decision between school finance equalization, on the one hand, and fiscal stability, on the other. For instance, school finance reforms (either through court order, legislation, or changes to the state's education funding formula) have been shown to reduce cross-district spending inequality among low- and high-property wealth districts (Candelaria and Shores, 2018; Jackson, Johnson, and Persico, 2016; Lafortune, Rothstein, and Schanzenbach, 2017; Sims, 2011; Steinberg, Quinn & Anglum, 2018). If such equalization efforts shift the responsibility for funding schools to the state level, and thereby make district spending more vulnerable to future recessionary events, then policymakers may be reluctant to pursue equalization.

However, the degree of centralization in a state's education finance system and a district's fiscal vulnerability following recessionary events may be correlated with omitted variables or confounded by pre-recession resource trends. For example, districts located in states with greater fiscal centralization may also have experienced different local labor market shocks following the onset of the Great Recession. If the recession was more severe in districts (nested within states) that were more reliant on state revenues, then the observed fiscal fallout may be due to recessionary intensity – the extent of employment loss in local communities – and not the degree of centralized education spending. Therefore, if post-recession shocks to local labor markets explain more of the variation in fiscal fallout than a state's relative contribution to education spending, then the policy response will likely be quite different. Instead of abandoning (or, rolling back) efforts to reduce cross-district spending inequality via state equalization efforts, more targeted approaches to distributing aid during recessionary periods should be pursued. Such targeted approaches would consider the magnitude of within-state variation in local labor market shocks (i.e., recession intensity) and seek to provide additional educational resources to those school districts [End Page 124] disproportionately affected by recessionary events.

While the American Recovery and Reinvestment Act (ARRA) of 2009 provided substantial federal fiscal stimulus to schools and districts nationwide – $97.4 billion in total, of which $48.6 billion was apportioned directly to state education systems – ARRA aid was not allocated to districts based on differences in, for example, spending declines following the Great Recession. Instead, ARRA funds were apportioned to state education systems based on the state's population share, and then allocated to districts within states based on a state's pre-existing education funding formula (Evans et al., 2017; Steinberg, Quinn & Anglum, 2018). Thus, while ARRA's ostensible purpose was to mitigate the shock of the recession on district education budgets, no empirical evidence exists as to whether the allocation of ARRA was effective at insulating districts most adversely impacted by the economic downturn of the Great Recession.

Our contributions in this paper are two-fold. First, we examine whether postrecession declines to district resources are explained by the degree of centralization in a state's education finance system or local labor market shocks. To do so, and in contrast to prior work (e.g., Evans, et al., 2017), we leverage populationlevel panel data and model school district spending for the 2002-03 through 2012-13 period in a difference-in-differences framework that accounts for district-specific secular trends. We find that the degree of state fiscal centralization is correlated with declines in state revenues, a result that is consistent with prior evidence on district resource shocks in the wake of the Great Recession (Evans et al., 2017; Jackson et al., 2018), Yet, we find that state fiscal centralization is uncorrelated with declines in district spending. Our findings indicate that recession-induced declines in spending are driven not by changes in state revenues, but instead by significant and substantive recession-induced declines in local revenues.

At the same time, recession-induced declines in total spending varied by district wealth. In states where state fiscal centralization was greatest, higher wealth districts were better able to offset spending losses by increasing local revenues compared to low wealth districts. Thus, low property wealth districts in states more reliant on state aid were especially vulnerable to the economic shock of the Great Recession because they were ill-equipped to offset state revenue declines with increases in local revenues. Given that local revenue declines were disproportionately concentrated among resource poor school districts, targeted stimulus aid would provide necessary fiscal support to districts most vulnerable to recession-induced spending losses – low wealth districts located in states with greater fiscal centralization who were exposed to the most severe employment shocks of the Great Recession.

Our second contribution is to examine whether federal fiscal stimulus (via [End Page 125] ARRA) was allocated to districts in a manner that mitigated spending declines. We show that ARRA did not mitigate – and, in some cases exacerbated – the spending gap between districts most and least adversely affected by recession-induced local labor market shocks. Specifically, ARRA aid was distributed almost equally (in terms of per-pupil allocations) across districts that experienced different local labor market shocks, a consequence of the fact that ARRA aid was not targeted to districts based on the extent of recession-induced spending declines. Thus, while districts with the most severe local labor market shocks had significantly greater declines in spending – $1,916 per pupil per year – than districts with the least severe labor market shocks –$877 per pupil per year – the provision of federal stimulus aid did little to offset these differential declines in school spending. Indeed, ARRA aid accounted for 24% of the loss in spending experienced by districts least affected by the economic downturn during the 2008-09 through 2012-13 period; among districts most adversely affected by the Great Recession, ARRA aid accounted for just 11% of the loss in spending. So, while the magnitude of ARRA aid varied little across districts, we find significant variation in the extent to which ARRA aid offset post-recession spending losses for districts differentially impacted by the economic consequences of the Great Recession.

Taken together, this paper provides new insights into the distribution of district revenue and expenditure declines following the onset of the Great Recession, the sources of these post-recession resource shocks, and the (potentially unintended) role that fiscal federalism played in widening district spending inequality in the wake of recessionary events. In the proceeding sections, we begin by describing the data and population of U.S. school districts included in our analytic sample. We then describe our approach for measuring the intensity of the recession (i.e., local labor market shocks) and the degree of fiscal centralization in a state's education finance system. Next, we present our empirical approach for estimating the relative influence of recession intensity and the degree of state fiscal centralization on district educational resources. Then, we describe how federal fiscal stimulus was distributed across districts, and whether federal aid mitigated any resource shocks associated with the Great Recession. We discuss our findings and conclude.

data & sample

We construct a district-level panel dataset consisting of the population of traditional public school districts in the continental United States for the 2002-03 through 2012-13 school years.1 We combine revenue and expenditure information [End Page 126] from the U.S. Department of Education's Common Core of Data (CCD) with economic, employment and housing information from the U.S. Department of Labor's Bureau of Labor Statistics (BLS), the U.S. Census and American Community Survey (ACS). Data from CCD includes district-by-year totals of ARRA-specific current and capital expenditures.2

District-level revenue data are from the CCD Local Education Agency Finance Survey (F-33), and include revenue from state, local and federal sources. District expenditure data also come from the CCD F-33 survey, and include total expenditures, instructional expenditures, and capital expenditures.3 Finally, in the 2008-09 through 2012-13 school years, the CCD F-33 survey reports two district-level expenditure measures derived from fiscal stimulus provided via ARRA funds – ARRA current expenditures and ARRA capital expenditures. We combine these two variables to measure total district-level ARRA spending for the 2008-09 to 2012-13 period. To generate a variable which measures district-level expenditures in the absence of ARRA aid, we subtract total ARRA expenditures from total district expenditures. We convert all revenue and expenditure variables to real ($2013) per-pupil dollars (using district enrollment totals) and eliminate outliers based on an algorithm akin to Murray, Evans and Schwab (1998) and Berry (2007).4

Measuring Recession Intensity

Following Ananat et al. (2017) and Yagan (2017), we measure recession intensity as the net change in log employment among all counties in the United States. Specifically, we measure recession intensity as:

inline graphic

where denotes the number of employed workers in county c in the Spring of academic year t,5 and where agg denotes total employment across the continental [End Page 127] U.S. in year t.6 Each county's recession intensity is measured as the change in log employment during the recessionary period (Spring 2007 to Spring 2010, or fiscal years 2007 to 2010) relative to the county's pre-recession trend (Spring 2003 to Spring 2006) and is then normalized by subtracting the aggregate employment trend. This measure of recession intensity accounts for counties' pre-recession secular trends in log employment. Notably, recession intensity is predominantly a within-state phenomenon – 76% of the variation in recession intensity occurs within states.7

Measuring State Fiscal Centralization

Following Evans et al. (2017) and Jackson et al. (2018), we quantify a state's relative investment in total education spending as:

inline graphic

where StateShares is equal to total state revenues across all d districts in state s in 2007-08 divided by total education revenues (i.e., revenues from state, local, federal and other sources) across all d districts in state s in 2007-08. StateShares measures the relative contribution to total revenues in state s just prior to the recession and the extent of state fiscal centralization.8

We generate three versions of Recessionc and StateShares. First, we convert the continuous measures Recessionc and StateShares into four quartiles. For example, quartile four of Recessionc includes districts (nested within counties) that experienced the largest net employment loss; quartile four for StateShares includes districts (nested within states) located in a state that contributes the largest share of total K-12 revenues.

Because Recessionc and StateShares are measured on different scales, we also convert them into two common metrics. For the first common metric, we standardize the variables to be mean zero and standard deviation one. Here, a 1-unit change is equivalent to a one standard deviation increase in either net job loss or state contribution to total revenues. For the second common metric, we convert the variables to rank-units (Chetty, et al., 2014). Here, a 1-unit change is equivalent [End Page 128] to a one rank-unit increase in either of the two measures.

Sample

Our analytic sample consists of 14,139 unique districts for the continental United States across the 2002-03 through 2012-13 school years, including 145,880 district-by-year observations.9 Table 1 summarizes district revenues and expenditures.10 On average, districts that experienced the greatest net job loss following the recession received significantly less total education revenue – nearly $2,000 per pupil less – than districts least affected by the economic recession. These districts, in turn, had fewer resources to dedicate to total and instructional education expenditures. Similarly, districts located in states that contributed the largest share of total K-12 revenues received significantly less total education revenue – approximately $1,250 per pupil less – than districts located in states that contributed the least toward education spending. Districts in states with the greatest fiscal centralization received approximately half of the local education revenue received by districts in states with the least fiscal centralization. However, part of this difference is made up by state revenues, as districts located in states with the greatest fiscal centralization received approximately $1200 more per pupil, on average, in state revenues than districts in states with the least fiscal centralization.

These patterns reveal that districts experiencing the greatest economic shock of the recession and located in states that contributed the most to education revenues spent, on average, the least on education during the study period (i.e., 2002-03 to 2012-13 school years). In districts with the greatest economic shock, this spending difference is attributable to differences in both local and state revenues. In districts (within states) with the greatest state revenue share, the spending difference is driven solely by differences in local revenues and is partially offset by differences in state revenues. We next describe the empirical approach we use to identify the predictors of post-recession changes in education revenues and expenditures.

empirical approach

We estimate the magnitude of district-level resource shocks in the wake of the Great Recession in two stages. In the first stage, we estimate the following model: [End Page 129]

Table 1. District Revenue and Expenditures, by Recession Intensity and State Investment in Education
Click for larger view
View full resolution
Table 1.

District Revenue and Expenditures, by Recession Intensity and State Investment in Education

inline graphic

In equation (3), Resourcesdt corresponds to a measure of: (i) per-pupil revenue (state, local or federal revenues) in district d in school year t; or (ii) per-pupil expenditures (total, instructional or capital expenditures) in district d in school year t (a total of six separate outcomes). The variable λd are district fixed effects and t is a continuous variable for the academic year, such that t=1 in the 2002-03 school year, 2 in the 2003-04 school year, up to the 2007-08 school year.

This first-stage equation purges estimates of district resource changes of unit-specific variation that is time-invariant and unit-specific time-varying secular trends prior to the onset of the Great Recession. From equation (3), we predict inline graphic for the entire period, i.e., 2002-03 through 2012-13 school years, effectively forecasting district-level resources throughout the recessionary period based on pre-recession district-level trends. The residuals from this prediction [End Page 130] inline graphic form the outcome variables used in the second stage (described below). Detrending the data in this way parallels our approach to defining and measuring Recessionc; that is, both recession intensity and our analytic strategy estimate post-recession resource changes net of pre-recession secular resource trends. Finally, using only pre-recession data to approximate the counterfactual addresses the problem of endogenous selection (or collider bias), which would occur if we included time-varying covariates (e.g., district demographic and economic characteristics) that are themselves associated with treatment (Elwert and Winship, 2014).

To estimate the change in district revenues and expenditures in the years following the onset of the Great Recession as a function of recession intensity and state fiscal centralization, we estimate the following three models:

inline graphic
inline graphic
inline graphic

In equations (4a) (4c), the variable Postt is equal to unity in the 2008-09 through 2012-13 school years and zero otherwise; the model excludes year fixed effects since inline graphic has been purged of pre-recession district-level secular trends. The interaction of Postt with measures of recession intensity and fiscal centralization (i.e., StateShare) enables an estimate of the average change in resources (relative to pre-recession trends) in the post-recession period as a function of either recession intensity or state investment in education, and εdt is a mean-zero random error term.

This specification is functionally equivalent to a detrended difference-indifferences approach. To see this, note that because we detrend each district's pre-recession outcome trends, estimates of βq from Equation (4a) correspond to the average net change in Resourcesdt in inline graphic (i.e., recession intensity quartile q) following the onset of the Great Recession (i.e., 2008-09 through 2012-13 school years), net of each district's pre-recession outcome slope. Similarly, from Equation (4b), estimates of γq correspond to the average net change in Resourcesdt in inline graphic (i.e., fiscal centralization quartile q) following the onset of the Great Recession, net of each district's pre-recession outcome slope. Thus, estimates of inline graphic and inline graphic comprise the first difference, and differencing βq=4 from βq=Q and λq=4 from λq=Q, where the subscript q is one of quartiles [End Page 131] q ∈ {1,2,3} comprise the second difference.11

In Equation (4c), we estimate marginal coefficients for β and γ, net of recession intensity and state revenue share. To reduce the dimensionality of the estimating equation, we replace the quartile indicators of Recessiond and StateShareq with their continuous counterparts, described above. Specifically, we estimate two separate equations in which recession intensity and state revenue share are standardized or converted to rank-units. Equation (4c) is akin to a horse-race model, testing whether district resource outcomes are differentially related to recession intensity and state fiscal centralization, net of the correlation between these two variables.

Following Abadie et al. (2017), we take an experimental design view to selecting the level of clustering. Specifically, since recession intensity status is assigned at the county level, clusters of districts (rather than individual districts) are assigned to a given level of recession intensity. The choice of clustering at the county level, rather than the state level, is further informed by Nichols & Shaffer (2007), who suggest that the data have at least 20 balanced clusters or 50 reasonably balanced clusters, and Rogers (1993), who suggests that no cluster contain more than 5% of the data. In our data, four states (California, Illinois, Texas and New York) each contain greater than 5% of the data, and a total of 26.3% of all district*year observations.12 Finally, because the data are district aggregates, we estimate weighted least squares regression with district enrollment as the weighting variable (Lafortune, Rothstein, and Schanzenbach, 2017; Solon, Haider, and Woolridge, 2015).13

Heterogeneity in Local District Response

Did districts that experienced similar economic shocks following the onset of the Great Recession fare better or worse in states with greater fiscal centralization? [End Page 132] Prior evidence suggests that school districts may respond differently to recessionary events depending on the severity of the recession and the degree of state centralization. First, districts stabilize educational revenues during recessionary periods by substituting local revenues (via increases in property tax rates) for declines in state support (Chakrabarti, Livingston & Roy, 2014; Dye & Reschovsky, 2008). In addition to the countercyclical role of local revenues, fiscal centralization may also provide opportunities for tax relief to local districts. For example, school finance reforms in New Hampshire and Pennsylvania increased fiscal centralization through the provision of equalization aid targeted to low-spending, property poor school districts. Following these reforms, approximately 90% of additional state grant aid was crowded out and used as local property tax relief (Lutz, 2010; Steinberg et al., 2016). Taken together, during recessionary events, districts located in states with greater fiscal centralization should be better able to raise additional local revenues (via increases in local tax rates) relative to districts similarly affected by the economic shock of the recession but located in states with less fiscal centralization.

With measures of both recession intensity and state fiscal centralization, we explore potentially important heterogeneity in district response to the magnitude of local labor market shocks. What follows are three empirical predictions. First, we expect education spending to increase more, on the margin, in districts experiencing more severe recession-induced local labor market shocks in states with greater fiscal centralization. Second, we expect these relative spending increases to be driven by increases in local revenues. Third, these relative spending increases will be greater in property wealthy districts, via increases in local revenues, than in districts with less property wealth. Thus, we expect the local subsidy resulting from the degree of state fiscal centralization to be disequalizing.

We examine these empirical predictions in the context of the following triple-interaction model:

inline graphic

The interaction term (Recessiond * StateShares * Postt) provides insight into the extent to which district resources changed in the post-recession period in districts most adversely affected by the recession (i.e., districts with the most net job loss) located in states with greater state fiscal centralization (i.e., states that contributed the largest relative share toward education spending). The model controls for recession intensity and fiscal centralization and estimates, on the margin, changes in spending in districts most affected by the recession [End Page 133] and located in states with greater state fiscal centralization. We examine the first empirical prediction by assessing estimates of ψ where inline graphic equals (detrended) per pupil total expenditures. If inline graphic, then spending will have increased more in districts in which the state's relative contribution to educational revenues is greater and where the recession was most severe. We examine the second prediction by estimating two regressions, alternatively replacing inline graphic with detrended per pupil local and state revenues, respectively. We predict that any relative increases in spending will be attributable to increases in local revenues and not to state revenues. To evaluate the third prediction, we generate measures of district housing values and household income from the 2000 decennial Census ACS and convert these measures into within states terciles.14 We then estimate Equation (5) for each of these property wealth terciles and compare estimates for inline graphic, where w indexes terciles of housing values and household income, respectively (i.e., w = 1 corresponds to low wealth districts and w = 3 corresponds to high wealth districts). If inline graphic, then high property wealth districts will have raised spending and local revenues relatively more than low property wealth districts, controlling for recession intensity and the degree of state fiscal centralization.

Examining the Role of Federal Fiscal Stimulus

Finally, we examine whether federal fiscal stimulus was allocated in relation to the magnitude of the economic shock of the Great Recession. To do this, we estimate a variant of Equation (4c). Specifically, we include two outcome variables for inline graphic described above: (i) total expenditures inline graphic; and (ii) total expenditures minus ARRA-specific expenditures inline graphic.15 Using a seemingly unrelated regression framework, we estimate the following system of equations:

inline graphic
inline graphic

In Equation (6a), the inline graphic coefficients describe total spending losses in each year following the Great Recession in inline graphic in years θt (i.e., 2008-09 [End Page 134] through 2012-13). In Equation (6b), the inline graphic coefficients describe total spending losses in the absence of ARRA in each year following the Great Recession. In Equations (6a) and (6b), we also replace θt with Postt to provide the average change in spending for the entire post-recession period (i.e., inline graphic becomes βq and inline graphic becomes λq). In both models, we control for the association between and spending loss by including interactions of year effects and state revenue share, where is standardized.16 Comparing estimates of inline graphic and estimates of inline graphic allow us to examine spending changes with and without federal stimulus for districts most and least affected by the recession, controlling for year-specific changes in spending correlated with state revenue contributions.

results

Our primary analysis is concerned with characterizing the relative contributions of state revenue share (i.e., the degree of fiscal centralization) and local labor market conditions (i.e., recession intensity) to fiscal fallout following the onset of the Great Recession. Our main finding is that the post-recession decline in district spending is highly correlated with recession intensity (see Figure 1 and Table 2, Panel A). For districts experiencing the largest net employment loss (i.e., recession intensity quartile 4), total expenditures declined annually by $1,917 per pupil, on average, across the post-recession period (i.e., 2008-09 through 2012-13 school years). Compared to districts least affected by the recession (i.e., recession intensity Q1), our difference-in-differences estimates suggest that the recession reduced total expenditures annually by $1,042 per pupil, on average, across the post-recession period for those districts most adversely affected by the recession (see Table 2, Panel A). Further, the recession's effect on total per pupil expenditure loss is monotonic (i.e., the Q4-Q1 difference is larger than the Q4-Q2 difference, which is, in turn, larger than the Q4-Q3 difference).

In contrast, we do not find any association between fiscal centralization and expenditure loss. The difference-in-difference estimates are not significantly different from zero for districts located in states that contributed the largest share of total K-12 revenues (i.e., fiscal centralization Q4) compared to districts in states that contributed the least and second least toward total K-12 revenues (i.e., fiscal centralization Q1 and Q2).

Second, we find that the recession was negatively related to capital spending but was not differentially related to instructional spending (see Figure 2). Among recession intensity Q4 districts, capital expenditures declined annually by $832 per pupil, on average, across the post-recession period. Compared to [End Page 135]

Figure 1 - No description available
Click for larger view
View full resolution
Figure 1.

recession intensity Q1 districts, our difference-in-differences estimates suggest that the recession reduced capital expenditures annually by $665 per pupil, on average, across the post-recession period (see Table 2, Panel A). In contrast, we see little to no difference among recession intensity quartiles in instructional spending declines. Thus, it appears that districts suffering the most from the economic shock of the recession substituted away from capital spending to smooth out instructional expenditures. This finding on the substitution away from capital to instructional expenditures among districts most adversely impacted by the economic shock of the recession is consistent with evidence from Jackson et al. (2018). Again, we find little evidence of a correlation between fiscal centralization and declines in either instructional or capital expenditures.

To what extent do the sources of district revenue change post-recession? Table 2 (Panel B) summarizes these results. We find that the recession was correlated with local revenue declines, with little consistent relationship with changes to revenues from state sources. Compared to recession intensity Q1 districts, our difference-in-differences estimates suggest that the recession was correlated with declines in local revenues annually of $588 per pupil, on average, across the post-recession period for those districts most adversely affected by the recession (see Table 2, Panel B). The recessionary effect on local revenues is also [End Page 136]

Table 2. Estimated Post-Recession Change in Expenditures and Revenues, by Quartile of Recession Intensity and State Investment in Education
Click for larger view
View full resolution
Table 2.

Estimated Post-Recession Change in Expenditures and Revenues, by Quartile of Recession Intensity and State Investment in Education

monotonic; the difference-in-differences estimate comparing Q4 to Q3 districts is nearly half the magnitude – $342 – of the estimated effect comparing Q4 to Q1 districts.

Not unexpectedly, fiscal centralization is associated with declines in state revenues, since districts more reliant on state aid realized fewer state revenues in the wake of the recession. However, relative declines in state revenue were approximately offset by relative gains in local revenues. The difference-in-differences estimate for state revenues (comparing fiscal centralization Q4 to Q1 districts) suggests that fiscal centralization is correlated with declines in state revenues of $772 per pupil annually. Conversely, the difference-in-differences estimate for local revenues (again comparing fiscal centralization Q4 to Q1) reveals that local revenues increased by $1133 per pupil annually in districts with the greatest fiscal centralization. These results suggest that districts in states with low fiscal [End Page 137]

Figure 2 - No description available
Click for larger view
View full resolution
Figure 2.

centralization– those states that contribute the least to education revenues – may be "maxed out" with respect to their local tax effort since they contribute a larger share toward education revenues. The consequence of these results is that local support of education may be more susceptible to recessionary shocks in states where the state contributes the least toward education. Note that these models do not control for recessionary intensity and state revenue share simultaneously, so we cannot separate the revenue losses due to fiscal centralization from those resulting from recession-induced local labor market shocks. To do so, we next present results which jointly condition on recession intensity and state fiscal centralization to further explore this empirical result.

Table 3 (Panel A) summarizes these results. We find that, conditional on fiscal centralization, recession intensity is highly correlated with expenditure loss. In contrast, controlling for recession intensity, we find no relationship between fiscal centralization and total or instructional expenditure declines. Specifically, a one standard deviation increase in recession intensity, net of fiscal centralization, is associated with a $518 per pupil annual decline in total expenditures across the recessionary period. Consistent with evidence presented in Table 2, results from Table 3, Panel A show that a one standard deviation increase in recession intensity is associated with a $333 per pupil annual decline in capital expenditures [End Page 138]

Table 3. Estimated Post-Recession Change in Expenditures and Revenues
Click for larger view
View full resolution
Table 3.

Estimated Post-Recession Change in Expenditures and Revenues

[End Page 139]

penditures across the recessionary period. We again find no association between recession intensity and instructional expenditures. These results are robust to the scaling of the explanatory variables (i.e., SD units or rank units).

Post-recession declines in district spending can further be explained by the differential change in revenue sources following the recession (see Table 3, Panel B). Specifically, while recession intensity is uncorrelated with state revenues, local revenues decline significantly, by $295 per pupil annually for each year after the Great Recession. In contrast, fiscal centralization is negatively correlated with state revenues (a one standard deviation increase in fiscal centralization is associated with a $315 per pupil annual decrease in state revenues) but positively predicts local revenues (a one standard deviation increase in fiscal centralization is associated with a $343 per pupil annual increase in local revenues). Thus, net of recession intensity, districts in states with greater fiscal centralization increased local revenues in support of education to offset declines in state revenues. Whether the recession stimulated local effort in support of education spending is a question we take up in the next section.

Heterogeneity in Local District Response

To what extent did local district resources vary in the post-recession period as a function of recession intensity and fiscal centralization? Our primary hypothesis is that the recession stimulated local effort in districts located in states that provide a relatively larger share of education revenues. Here, we present results from three empirical predictions that lend insight into this hypothesis.

First, we expect district spending to increase among districts more adversely affected by the recession (i.e., greater recession intensity) and located in states with greater fiscal centralization. If local revenues are counter-cyclical and substitute for state support during recessionary periods (Chakrabarti, Livingston & Roy, 2014; Dye & Reschovsky, 2008) and state aid subsidizes local tax effort in support of education spending (Lutz, 2010; Steinberg et al., 2016), then districts most adversely affected by the recession in states with greater fiscal centralization should increase education spending, conditional on recession intensity and the extent of fiscal centralization. This is indeed what we find. Districts that jointly have one standard deviation greater recession intensity and one standard deviation greater fiscal centralization, controlling for baseline levels of these variables, realized average annual increases in education spending of $260 per pupil across the post-recession period (see Table 3, Panel A, the coefficient on Recession Intensity*Fiscal Centralization). This result is robust to the functional form of the explanatory variables (i.e. SD units or rank units).

Second, we expect increases in educational expenditures to be driven by increases in local revenues. Again, if a state's (pre-recession) share of education [End Page 140] revenues subsidizes local spending and the onset of the recession stimulates a revenue generating effort on behalf of local school districts (i.e., the countercyclicality of local revenues), we would expect spending increases to be driven by increases in local revenues. Indeed, we find that most of the spending increase – approximately 54% – is due to increases in local revenues. Districts with one standard deviation greater recession intensity and one standard deviation greater fiscal centralization (controlling for baseline levels of these variables) increased average annual revenues from local sources by, on average, $141 per pupil across the post-recession period (see Table 3, Panel B, the coefficient on Recession Intensity*Fiscal Centralization). In contrast, state revenues are not statistically significantly correlated with the interaction between recession intensity and fiscal centralization.

Third, the consequences of a state's subsidy for local effort in support of educational spending during recessionary periods will be disequalizing. Higher wealth districts – those with greater median home prices and higher household incomes – will be more equipped to increase spending via increases in local revenues than resource poor districts, holding constant recession intensity and state fiscal centralization. Results summarized in Table 4 support this prediction. Namely, wealthier districts – those with either higher mean housing values or higher mean household income – generate significantly greater expenditures in the post-recession period than their resource-poor counterparts, between $381 and $430 per pupil annually (see Table 4, Panel A, DD Estimate), depending on the measure of wealth used. Moreover, these differential expenditure increases are funded in large part by differential increases in local revenues, between $168 and $215 per pupil annually (see Table 4, Panel B, DD Estimate).

The Role of Federal Stimulus in Post-Recession Spending Loss

In this section, we examine the extent to which federal fiscal stimulus, funded via the American Recovery and Reinvestment Act (ARRA), mitigated (or even exacerbated) the gap in spending loss between districts differentially exposed to net job loss following the onset of the Great Recession.

Table 5 summarizes estimates of the extent of spending loss funded by ARRA-specific expenditures across recession intensity quartiles. Figure 3 presents estimates of the post-recession spending loss, by recession quartile, with and without ARRA-supported expenditures. Table 5 (Panel A) displays regression estimates of ARRA expenditures by recession intensity quartile (i.e., from Equations (6a) and (6b)).We find that districts least affected by the recession spent slightly more ARRA-supported funds compared to districts most affected. Specifically, districts most adversely affected by the recession (i.e., Quartile 4) spent, on average, $226 per pupil annually in ARRA aid, whereas districts least affected [End Page 141]

Table 4. Estimated Post-Recession Change in Total Expenditures and Local Revenues, by Household Wealth
Click for larger view
View full resolution
Table 4.

Estimated Post-Recession Change in Total Expenditures and Local Revenues, by Household Wealth

by the recession (i.e., Quartile 1) spent, on average, $277 per pupil annually in ARRA-supported expenditures across the post-recession period.

Table 5 (Panel B) shows the ratio of ARRA expenditures relative to expenditure declines (in absolute values) in the post-recession period (i.e., inline graphic). Here we find meaningful differences across recession intensity quartiles. For example, across all post-recession years (i.e., 2008-09 through 2012-13), ARRA-specific expenditures accounted for 24% of total spending loss among districts least impacted by the Great Recession (i.e., Q1 districts). In contrast, ARRA-specific expenditures accounted for just 11% of total spending loss among districts most impacted by the Great Recession (i.e., Q4 districts). This 13 percentage-point difference in the ratio of ARRA-specific expenditure to post-recession spending loss is highly significant (see Table 5, Panel C). However, this difference is likely understated, since the overwhelming share of ARRA aid (on a per-pupil basis) was distributed in just two years – 2009-10 and 2010-11 (see Table 5, Panel A). [End Page 142]

inline graphic
Table 5. Estimated Contribution of ARRA to Spending Loss, by Recession Intensity
Click for larger view
View full resolution
Table 5.

Estimated Contribution of ARRA to Spending Loss, by Recession Intensity

In 2009-10, ARRA-specific expenditures accounted for 87% of total spending loss among districts least impacted by the Great Recession (i.e., Q1 districts), but only 30% of total spending loss among districts most impacted by the Great Recession (i.e., Q4 districts); in this way, ARRA aid was almost fully compensatory for districts least affected by the economic recession (see Table 5, Panel B).

These results suggest two key findings with respect to the role that federal fiscal stimulus played in the wake of the Great Recession. First, while there were few substantive differences across quartiles of recession intensity in the magnitude [End Page 143]

Figure 3 - No description available
Click for larger view
View full resolution
Figure 3.

of ARRA expenditures (see Table 5, Panel A), we find significant differences in the extent to which ARRA aid offset post-recession spending losses across districts differentially impacted by the economic shock of the Great Recession. And second, ARRA aid was not distributed to school districts in a way that insulated districts from the recession's effect on education spending; instead, the distribution of ARRA aid, at best, did little to reduce the spending gap between districts and, at worst, exacerbated those gaps (see Figure 3).

conclusion

In this paper, we examine the correlates of district spending and revenue shocks following the onset of the Great Recession and the role of fiscal federalism in mitigating these shocks. Our findings point to the deleterious effect of recessionary events on district resources and spending, while revealing a more nuanced role for state fiscal centralization in explaining district resource shocks. Specifically, we find that, on average, recession intensity is correlated with declines in both local revenues and total spending. On the other hand, state fiscal centralization is correlated with declines in state revenues but not declines in total spending, as state revenue losses were, on average, offset by increases in local revenues. However, this offset is disproportionately distributed among high [End Page 144] property wealth districts that were able to compensate for declines in state revenues while low property wealth districts were not. Thus, low-wealth districts in states with greater fiscal centralization were especially vulnerable to recessionary events and would stand to benefit from targeted aid. ARRA aid, however, was untargeted and therefore did not mitigate the disproportionate declines in total spending following the onset of the Great Recession.

Findings from this paper extend prior evidence on the correlates of post-recession declines in district resources (Evans et al., 2017). While our measure of state fiscal centralization is consistent with Evans et al. (2017), our analytic strategy departs from theirs in important ways. First, our measure of recession intensity accounts for pre-recession secular trends in employment, while Evans et al. (2017) leverage only post-recession declines in employment to capture recession intensity. Second, our econometric approach leverages the panel nature of these data by accounting for pre-recession secular trends in district spending; absent this, post-recession changes in spending may be confounded by prerecession trends. By conditioning on pre-recession trends in both employment and district resources, we find that district spending declined primarily because of recession-induced local labor market shocks, and not because of the degree of fiscal centralization in a state's education finance system.

These results have important policy implications, especially since another economic recession is certain. If the degree of centralization in a state's education finance system—which has increased, in part, due to efforts to equalize school resources among property poor and rich districts—makes district spending more vulnerable to recessionary events, then policymakers may (rightly) reduce efforts to promote equality and, instead, shift the responsibility for funding schools back to school districts. Because districts can respond more quickly to changes in economic conditions than can states, a shift away from state to local aid in support of funding schools would better insulate school district finances during fiscal downturns. However, if the degree of state fiscal centralization is high, districts may be able to raise local revenues (via property tax increases) in times of fiscal duress to offset state revenue declines. Thus, our findings indicate that policy efforts designed to equalize cross-district, within-state differences in spending need not be abandoned. Instead, state and federal policy can be used to provide targeted aid to districts most affected by fiscal downturns – low-wealth districts heavily reliant on state revenues that are, at the same time, limited in their ability to raise local revenues to offset state revenues declines. Yet, our findings also reveal that federal fiscal stimulus following the onset of the Great Recession did little to mitigate the differential fallout resulting from the Great Recession, and, in some cases, exacerbated differences in spending loss. [End Page 145]

Notably, the results presented in this paper are descriptive, even as we have employed econometric methods – such as a difference-in-differences framework – typically used for causal analysis. At the same time, these methods have utility in a descriptive context. For example, by detrending the data, we adjust for heterogeneity in each district's resource trends in the years leading up to the onset of the Great Recession. Moreover, the inclusion of district fixed effects adjusts for time-invariant characteristics of districts that may be correlated with spending preferences. And, the difference-in-differences framework allows us to compare changes in spending and revenue patterns between districts more/less reliant on state aid and more/less affected by the Great Recession.

Yet, we acknowledge that this approach does not fully address district tax and spending decisions that may be endogenous and reflect time-varying unobserved processes that may be correlated with the onset of the Great Recession. For instance, states with greater fiscal centralization may also prioritize education relative to other state expenditures, and this budget priority may affect the state's fiscal response during recessionary events by, for example, shielding education from fiscal cuts relative to other state expenditures. Further, state aid may also respond to local effort, meaning that states may reduce educational aid if they observe increased local effort by districts and municipalities that increase their property tax rates. Thus, the variables state fiscal centralization and recessionary intensity should not be interpreted causally, as changes in observed spending that correlate with these variables may also reflect unobserved endogenous responses.

Finally, states and districts vary in more ways than recessionary intensity and state fiscal centralization. Specifically, state and local responses to fiscal shocks may interact with the state's school funding formula, as different school equalizations create different incentives for local school districts and affect tax prices for local revenues (e.g., Hoxby, 2001). Insofar as a state's funding formula is a fixed characteristic of the state, our econometric approach adjusts for these differences among states. However, more work should be done to better understand whether differences in state funding formulas drive heterogeneity in district spending declines, especially given the extent of heterogeneity in the local revenue responses during the recession.

disclosures

The authors have received financial support from the Russell Sage Foundation and the W.T. Grant Foundation. The authors have no financial arrangements that might give rise to conflicts of interest with respect to the research reported in this paper. [End Page 146]

Kenneth Shores

Kenneth Shores is Assistant Professor in Human Development and Family Studies at Pennsylvania State University.

Matthew P. Steinberg

Matthew P. Steinberg is Associate Professor of Education Policy at George Mason University.

references

Abadie, A., Athey, S., Imbens, G. W., & Wooldridge, J. (2017). When Should you Adjust Standard Errors for Clustering? (No. w24003). National Bureau of Economic Research.
Ananat, E. O., Gassman-Pines, A., Francis, D.V., & Gibson-Davis, C.M. (2017). "Linking Job Loss, Inequality, Mental Health, and Education." Science, 356(6343): 1127-1128.
Berry, C. (2007). "The Impact of School Finance Judgments on State Fiscal Policy." School Money Trials: The Legal Pursuit of Educational Adequacy, 213-242.
Business Cycle Dating Committee, National Bureau of Economic Research (September 20, 2010). http://www.nber.org/cycles/sept2010.html
Cameron, A. C., & Miller, D. L. (2015). "A practitioner's Guide to Cluster-Robust Inference." Journal of Human Resources, 50(2), 317-372.
Candelaria, C.A., and Shores, K.A. (2017). "Court-ordered Finance Reforms in the Adequacy Era: Heterogeneous Causal Effects and Sensitivity." Education Finance and Policy (2017): 1-91.
Chakrabarti, R., & Livingston, M. (2013). "The Long Road to Recovery: New York Schools in the Aftermath of the Great Recession." FRB of New York Staff Report, (631).
Chakrabarti, R., Livingston, M., & Roy, J. (2014). "Did Cuts in State Aid during the Great Recession Lead to Changes in Local Property Taxes?" Education Finance and Policy, 9(4), 383-416.
Chetty, Raj, et al. (2014). "Where is the Land of Opportunity? The Geography of Intergenerational Mobility in the United States." The Quarterly Journal of Economics 129(4): 1553-1623.
Dye, R. F., & Reschovsky, A. (2008). "Property Tax Responses to State Aid Cuts in the Recent Fiscal Crisis." Public Budgeting & Finance, 28(2), 87-111.
Elwert, F, & Winship, C. (2014) "Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable." Annual Review of Sociology, 40: 31-53.
Evans, W. N., Schwab, R. M., & Wagner, K. L. (2017). "The Great Recession and Public Education." Education Finance and Policy.
Ferman, Bruno, and Cristine Pinto. (2016). "Inference in Differences-in-Differences with Few Treated Groups and Heteroskedasticity." Retrieved from http://economics.ucdavis.edu/events/papers/1023Ferman.pdf.
Ferman, Bruno, and Cristine Pinto. (2018) "Inference in Differences-in-Differences with Few Treated Groups and Heteroskedasticity." The Review of Economics and Statistics. Doi: https://doi.org/10.1162/rest_a_00759
Hoxby, C. M. (2001). "All School Finance Equalizations are Not Created Equal." The Quarterly Journal of Economics, 116(4), 1189-1231.
Hurd, M. D., & Rohwedder, S. (2010). "Effects of the Financial Crisis and Great Recession on American Households" (No. w16407). National Bureau of Economic Research.
Jackson, C. K., Johnson, R. C., & Persico, C. (2015). "The Effects of School Spending on Educational and Economic Outcomes: Evidence from School Finance Feforms." The Quarterly Journal of Economics, 131(1), 157-218.
Jackson, K., Wigger, C., & Xiong, Heyu. (2018). "Do School Spending Cuts Matter? Evidence from the Great Recession." NBER Working Paper 24203.
Lafortune, J., Rothstein, J. & Schanzenbach, D. (2018). School finance reform and the distribution of student achievement. American Economic Journal: Applied Economics, 10(2): 1-26.
Leachman, M., & Mai, C. (2014). "Most States Still Funding Schools less than before the Recession." Center on Budget and Policy Priorities.
Lutz, B. (2010). "Taxation with Representation: Intergovernmental Grants in a Plebiscite Democracy." Review of Economics and Statistics, 92(2), 316-332.
Murray, S. E., Evans, W. N., & Schwab, R. M. (1998). "Education-finance Reform and the Distribution of Education Resources." American Economic Review, 789-812.
Nichols, A., & Schaffer, M. E. (2007, September). "Clustered Standard Errors in Stata." In United Kingdom Stata Users' Group Meetings 2007 (No. 07). Stata Users Group.
Rogers, W. (1993). "Regression Standard Errors in Clustered Samples." Stata Technical Bulletin, 13, 19-23.
Sims, D.P. (2011). "Lifting all Boats? Finance litigation, Education Resources, and Student Needs in the Post-Rose Era." Education Finance and Policy 6.4 (2011): 455-485.
Solon, G., Haider, S.J., & Wooldridge, J.M. (2015). "What are we Weighting For?" Journal of Human Resources, 50(2): 301-316.
Steinberg, M., Quinn, R., & Anglum, C. (2018). "Education Finance Reform and the Great Recession: Did State Policy and Fiscal Federalism Improve Education Spending, School Resources and Student Achievement in Pennsylvania?" Unpublished Manuscript.
Steinberg, M., Quinn, R., Kreisman, D., & Anglum, C. (2016). "Did Pennsylvania's Statewide School Finance Reform Increase Education Spending or Provide Tax Relief?" National Tax Journal, 69(3), 545-582.
Wolfers, J. (2006). "Did Unilateral Divorce Laws Raise Divorce Rates? A Reconciliation and New Results." American Economic Review 96(5): 1802-1820.
Wolff, E., Owens, L. A., & Burak, E. (2011). "How Much Wealth was Destroyed in the Great Recession." In D. B. Grusky, B. Western, & C. Wimer (Eds.), The Great Recession (pp. 127-158). New York: Russell Sage Foundation.
Yagan, D. (2017). "Employment Hysteresis from the Great Recession." NBER Working Paper 23844.

Footnotes

1. Data for Washington, DC, Hawaii and Alaska are excluded, as are districts comprised only of charter schools.

2. For measures of district resources (revenues and expenditures), we consider the 2002-03 through 2007-08 school years as pre-recession years; the 2008-09 through 2012-13 school years are the post-recession years. Though the Great Recession officially began in December 2007, state (and district) education budgets for the 2007-08 academic year are determined prior to the start of the academic year (i.e., September 2007), and therefore prior to the recession's onset.

3. Instructional expenditures include payments from all district funds for salaries, employee benefits, supplies, materials, and contractual services for elementary/secondary instruction; excludes capital outlay, debt service, and interfund transfers for elementary/secondary instruction. Capital expenditures include expenditures for construction of buildings, roads, and other improvements, and for purchases of equipment, land, and existing structures.

4. Specifically, the algorithm calculates the average spending in each state and year and eliminates district values in that state-year which are less than 25% of the bottom 5th percentile or greater than 200% of the top 95th percentile.

5. For example, the Spring academic year 2010 corresponds to the academic year 2009-10, which in turn corresponds to fiscal year 2010.

6. Annual employment data are taken from the Quarterly Census of Employment and Wages (QCEW).

7. The variable will contain multiple districts d for every county c, as there are 3,051 counties in the continental U.S. for which we have school finance and employment data. The regression equation, described below, will include the subscript d for this reason, but recession intensity varies only across counties c.

8. Although the Great Recession officially began in December 2007, state (and district) education budgets for the 2007-08 academic year are determined prior to the start of the academic year (i.e., September 2007). Therefore, education revenues and spending in the 2007-08 school year will not be affected by the official onset of the Great Recession.

9. Districts need not be present for each year in the sample. In no individual year is the number of districts greater than 14,000. The average number of districts by year is 13,267 and ranges from 12,942 in the 2012-13 school year to 13,772 in the 2002-03 school year. For 2012-13, the NCES reports 13,515 school public school districts in the United States (including Washington DC, Hawaii and Alaska); our sample therefore includes 95.7% of the population of all U.S. school districts.

10. Note that recession intensity is measured within states while fiscal centralization (i.e., state share) is measured across states; i.e., recession intensity varies across districts (nested within counties), while fiscal centralization varies across states (but is constant across districts within states).

11. Note that if we include the terms λd and λdt (from equation (3)) in Equations (4a-4c) and replace inline graphic with Resourcesdt, estimates of βq and λq will be identical insofar as we estimate each of the post-recession effects non-parametrically by replacing Postt with year indicators θt for years 2008-09 to 2012-13. By estimating more traditional effect sizes that replace θt with a single indicator variable Postt, including unit-specific time trends is not equivalent to estimating on the residuals and as result estimates will be biased. See Wolfers (2006) for discussion and explanation in the context of unilateral divorce laws.

12. Abadie et al. (2017) note that clustering at the most aggregate level (such as states), which has been recommended by applied economists (see e.g., Cameron & Miller (2015)), may "lead to standard errors that are unnecessarily conservative, even in large samples" (page 2).

13. Ferman and Pinto (2016) discuss the problem of inference in cases where heteroskedasticity is induced from the use of aggregate data. Specifically, they note that difference-in-differences with individual and aggregate data are identical insofar as "the estimator with aggregated data [uses] the number of observations per group x time cell as sampling weights," (p. 6). In our case, the sampling weights correspond to K-12 enrollment. (Note that this exact quotation is from the 2016 working paper and the forthcoming version in the Review of Economics and Statistics says only "[h]owever, both the diagnosis of the inference problem with existing methods and the solutions we propose are valid whether we have aggregate or individual-level data," (2019, p.8).)

14. Specifically, we use two variables from the 2000 decennial Census from the American Community Survey (ACS) Education Demographic and Geographic Estimates (EDGE). The variables are median home values and median family income. We construct within-state terciles for each of the two measures.

15. The amount of total ARRA aid per district is recorded in the NCES Common Core of Data F-33 file. We subtract total ARRA expenditures from total expenditures to recover spending in the absence of ARRA.

16. Results are robust to replacing the standardized version of StateShares with the rank-units (and are available from the authors upon request).

Additional Information

ISSN
1944-6470
Print ISSN
0098-9495
Pages
123-148
Launched on MUSE
2020-03-16
Open Access
No
Back To Top

This website uses cookies to ensure you get the best experience on our website. Without cookies your experience may not be seamless.