Abstract

We examine the impact of the Great Recession on public education finance and employment. Five major themes emerge from our work. First, nearly 300,000 school employees lost their jobs. Second, schools that were heavily dependent financially on state governments were particularly vulnerable to the recession. Third, local revenues from the property tax actually increased during the recession, primarily because millage rates rose in response to declining property values. Fourth, inequality in school spending rose sharply during the Great Recession. We argue, however, that we need to be very cautious about this result. School spending inequality has risen steadily since 2000; the trend in inequality we see in the 2008–13 period is very similar to the trend we see in the 2000–08 period. Fifth, the federal government's efforts to shield education from some of the worst effects of the recession achieved their major goal.

1.  Introduction

The recession that began in December 2007 was the most severe economic downturn in the United States since the Great Depression. The unemployment rate reached 10 percent in October 2009.1 Over eight million private sector jobs were lost and private employment did not return to pre-recession levels until spring 2014.2 As late as April 2016, there were over two million long-term unemployed people in the United States.3 Analysts often call this period the Great Recession (GR), a term that is well-deserved.

In this paper, we look at the impact of the recession on public schools. Our goal is to describe what happened to K–12 public education during the recession and to learn what we can about how to shield schools and their students from the worst effects of any future recessions.

Other papers have examined the impact of the GR on education. Several have focused on New York and New Jersey schools. Bhalla, Chakrabarti, and Livingston (2017) show that New York schools fared much better than New Jersey schools. New York received more than twice as much federal aid per pupil than New Jersey, and New Jersey state aid to schools fell much more sharply. Chakrabarti, Livingston, and Setren (2015) found that, in New York, increases in federal aid largely offset decreases in state aid, and consequently the recession had little impact on the overall level of education spending in the state. They also showed that the recession hit wealthy school districts the hardest. Chakrabarti and Sutherland (2013) argue that in New Jersey, noninstructional expenditures declined much more than instructional expenditures, nontenured teachers were more likely to be laid off, and expenditures in high poverty and urban school districts fell significantly.

In this paper, rather than examining the experience of a few states, we take a somewhat broader perspective. In section 2 we present a brief overview of the structure of education finance in the United States and then look at the aggregate effect of the recession on state and local government revenue, employment in public schools, and school spending. We compare the impact of the most recent recession to past recessions. In section 3 we use a balanced panel of district-level data to determine how the recession affected different types of schools. So, for example, we ask whether the structure of school finance was an important determinant of the impact of the recession. In section 4, we continue our work with the panel of districts and look at inequality in school spending from 1972 through 2013. An important question here is whether the recession had a particularly severe effect on schools that served children from low-income families. In section 5 we look at the efficacy of the federal government's efforts to offset at least part of the effect of the GR on public education. Section 6 includes a brief summary and conclusions.

Five major themes emerge from our work. First, the impact of the GR was unprecedented. Nearly 300,000 school employees lost their jobs, wiping out the aggregate gains made in reducing class size during the thirteen years before the recession. It took five years for state and local revenues to return to the pre-recession levels.

Second, schools in states where districts were heavily dependent on funds from state governments were particularly vulnerable to the effects of the GR. There has been a marked shift toward state-financed public schools over the past forty years, in part as a result of litigation and legislation to equalize school resources across districts. We show that revenues from the major state taxes—income and sales taxes—fell sharply over the GR. Our results suggest that an unintended side effect of these efforts has been to make school spending more vulnerable to recessions.

Third, despite the fact that the recession occurred at a time when property values were plummeting, property tax revenue, the mainstay of local school finance, actually rose over the course of the recession. Many school districts were able to offset the shrinking property tax base by raising the property tax rate. Property tax rates decline relatively little when property values increase and increase markedly when values decline, meaning the property tax is a stable source of revenue.

Fourth, inequality in school spending rose dramatically during the GR. School spending inequality had risen steadily since 2000 and the trend in inequality in the 2008 to 2013 period is similar to the trend we see in the 2000 to 2008 period. Thus, while the gap in spending between wealthy and poor schools rose during the recession, the role of the recession is much less clear.

Fifth, we argue that the federal government's efforts to shield education from some of the worst effects of the GR achieved their major goal. The State Fiscal Stabilization Fund (SFSF), created under the American Recovery and Reinvestment Act of 2009, provided $53.6 billion of funding for public schools during the early parts of the recession. We find that as a result of SFSF money, school spending was flat during the 2008–09 and 2009–10 school years.

2.  The Effect of the Great Recession on Education at the National Level

In this section, we present estimates of education finance, spending, and employment at the national level over the great recession. As we show later in this section, the impact of the great recession was in part a function of the way education finance has evolved over time. To set the stage for this discussion, we initially present some basic facts about K–12 education finance.

Some Basics of K-12 Education Finance

Annual real education spending per student nearly tripled between the 1970–71 and the 2012–13 school years.4 Education spending, however, is sensitive to the business cycle. In figure 1 we graph the de-trended residuals of real per student current expenditures versus the national unemployment rate.5 The graph shows a strong negative relationship between these two series except for the mid 1990s.6,7 Given this pattern, it is not a surprise to find a large drop in real current expenditures at the start of the GR.

Figure 1.

Time Series Plot of National Unemployment Rate and Detrended Real Current Expenditures per Pupil (2013$)

Figure 1.

Time Series Plot of National Unemployment Rate and Detrended Real Current Expenditures per Pupil (2013$)

There have been some significant changes in the way schools are financed in the United States over the last forty years. In figure 2 we summarize education revenues by source over time.8 State governments now play a much larger role in education finance than they once did. In the early part of the 20th century, nearly 80 percent of the revenues for public education came from local governments (Snyder 1993). In 1970, local governments provided 52.4 percent of K–12 revenues, and the state share was less than 40 percent. By 2008 the local share had fallen to 43.5 percent, and the state share had risen to 48.3 percent.

Figure 2.

Source of Revenues for K–12 Education, 1970—71 to 2011—12 School Years

Figure 2.

Source of Revenues for K–12 Education, 1970—71 to 2011—12 School Years

The growing role of the states in education is in part a response to a long series of court cases that have challenged the constitutionality of an education finance system that has led to wide disparities in education spending across school districts.9Serrano I in 1971 and subsequent cases led to a requirement of equal spending per student in California. More recent cases have been driven by concerns over the adequacy of funding for public education, in particular the funding of education for students from disadvantaged backgrounds. At last count, litigants had challenged the constitutionality of state school finance systems in forty-five states (Corcoran and Evans 2015). In most cases, a decision by a high court to overturn a state education financing system has been accompanied by a direct order to make fundamental changes to school funding formulas. State legislatures have also initiated their own far-reaching reforms to school finance systems in the wake of unsuccessful litigation (e.g., Georgia and Idaho), under the threat of litigation (e.g., Missouri and Oklahoma; see Minorini and Sugarman 1999), or in response to political pressure (e.g., Michigan).

Historically, the federal government has played a small role in K–12 education finance. The average federal share over the 1970–2008 period was 7.8 percent. The federal government did provide significant additional funding at the start of the GR in response to falling state and local tax revenues. As a consequence, in 2010 the federal share of education spending reached 13 percent. In the last several years the federal role has moved back toward historical levels.

In table 1 we report revenues to state and local governments from broad sources in 2012.10 In the first two columns we show the tax type and source. In the next two columns we report total revenues by type and the fraction of these total state revenues from this tax. We then present the same information for local governments, and then data for these two levels combined.

Table 1.
State and Local Tax Revenues by Source, 2012 (in Millions of Dollars)
StateLocalState + Local
Tax typeTax sourceTotal $Share %Total $Share %Total $Share %
Income Individual 225,019 36.5 84,839 10.8 309,858 22.1 
 Corporate 33,304 5.4 15,929 2.0 49,233 3.5 
Property  10,233 1.7 464,167 59.3 474,400 33.9 
Sales General 185,043 30.0 111,224 14.2 296,267 21.2 
 Motor fuels 31,357 5.1 11,173 1.4 42,530 3.0 
 Tobacco 13,250 2.2 4,079 0.5 17,329 1.2 
 Alcohol 4,516 0.7 1,953 0.2 6,469 0.5 
Motor vehicle license  18,820 3.1 9,933 1.3 28,753 2.1 
Other taxes  94,658 15.4 79,743 10.2 174,401 12.5 
Total  $616,200  $783,040  $1,399,240  
StateLocalState + Local
Tax typeTax sourceTotal $Share %Total $Share %Total $Share %
Income Individual 225,019 36.5 84,839 10.8 309,858 22.1 
 Corporate 33,304 5.4 15,929 2.0 49,233 3.5 
Property  10,233 1.7 464,167 59.3 474,400 33.9 
Sales General 185,043 30.0 111,224 14.2 296,267 21.2 
 Motor fuels 31,357 5.1 11,173 1.4 42,530 3.0 
 Tobacco 13,250 2.2 4,079 0.5 17,329 1.2 
 Alcohol 4,516 0.7 1,953 0.2 6,469 0.5 
Motor vehicle license  18,820 3.1 9,933 1.3 28,753 2.1 
Other taxes  94,658 15.4 79,743 10.2 174,401 12.5 
Total  $616,200  $783,040  $1,399,240  

Source: Quarterly Summery of State and Local Taxes, U.S. Census Bureau.

There is significant variation in state tax structures. Seven states have no income tax (and two others tax only dividend and interest income); five states do not have a sales tax.11 Despite this heterogeneity, the income and sales taxes are key sources of revenue for the states. Averaging across all states, 42 percent of state government revenues come from individual and corporate income taxes and an additional 37 percent come from various sales taxes. Less than 2 percent of state revenues come from property taxes. We find a very different picture at the local level. The property tax generates about 59 percent of local government tax revenue whereas sales and income taxes produce about one fifth of total revenues. Given these facts, one might then expect that the collapse of the housing markets at the start of the GR might have had a particularly severe impact on schools. We consider this issue in the next section of the paper.

The Impact of the Great Recession on the Financing of Public K–12 Education

This section of the paper focuses on the effects of the GR on the financing of public K–12 education.12 The impact of the GR on state and local tax revenues was dramatic. In figure 3, we present a four-quarter moving average index of real state and local tax revenue from the four largest sources of revenues (property taxes, income taxes, sales taxes, and corporate income taxes).13 We present data for the GR (which began in the fourth quarter of 2007) and the two previous recessions (which began in the third quarter of 1990 and the first quarter of 2001).14 We set revenues equal to 100 at the start of a particular recession. The horizontal axis shows the number of quarters after the start of the recession.

Figure 3.

Index of Four-Quarter Moving Average of Quarterly State and Local Tax Revenues from Property, Sales, Income, and Corporate Taxes over the Last Three Recessions (Start of Recession = 100)

Figure 3.

Index of Four-Quarter Moving Average of Quarterly State and Local Tax Revenues from Property, Sales, Income, and Corporate Taxes over the Last Three Recessions (Start of Recession = 100)

Figure 3 shows that the effect of the GR on state and local tax revenue was unprecedented. State and local revenues were constant for about a year after the start of the recession but then quickly fell by about 5 percent. Revenues remained flat for five quarters and then rose very slowly. It was not until eighteen quarters after the start of the recession that state and local tax revenues returned to pre-recession levels. But of course, the demand for state and local government services was far from flat during this period. For example, Medicaid rolls grew by 11.8 million people—an increase of 28 percent—between 2007 and 2012.15

We see a very different story when we look at previous recessions. Revenues never fell during the 1990 recession. Real revenues were 8 percent higher eleven quarters after the start of the recession and then remained flat for the next four quarters. In the 2001 recession, revenues fell for nine quarters then increased dramatically. As we will say many times in this paper, the impact of the GR was very different from previous recessions.

Figure 4 looks at the time path of an index of a four-quarter moving average of major sources of real state and local tax revenues after the start of the GR. All of the indexes are set to 100 at the start of the recession. Revenues from state and local income taxes, sales taxes, and corporate income taxes all fell very sharply at the start of the recession. Individual income tax collections were down 16 percent eight quarters into the recession and remained 10 percent below pre-recession levels for thirteen quarters. Income tax revenues were still 2 percent below its 2007 levels twenty quarters later. Sales tax revenues declined more slowly than income tax revenues but these revenues were 8 percent below the 2007 level fifteen months later. Revenues from corporate income taxes reached a nadir eleven quarters after the start the GR and were down 28 percent. Corporate income and sales taxes were still lower by 22 and 4 percent, respectively, five years after the start of the GR.

Figure 4.

Index of Four-Quarter Moving Average of Sources of State and Local Tax Revenues over the Great recession (Start of the Recession = 100)

Figure 4.

Index of Four-Quarter Moving Average of Sources of State and Local Tax Revenues over the Great recession (Start of the Recession = 100)

Property taxes followed a very different pattern. As figure 4 shows, revenues from property taxes actually grew steadily during the first three years of the recession. They then fell slightly but remained 10 percent above pre-recession levels fifteen quarters later. This is in some ways a surprising result. The housing market collapse was a key element of the GR. The Case-Shiller Home Price Index, a leading measure of housing prices, suggests that the housing bubble began to deflate at least two years before the recession.16 By the fourth quarter of 2007 the average price of a home was 20 percent below its 2005 peak. Prices continued to fall during the first year of the recession, and by December 2008 the real price of a home was roughly one third below its peak. New home starts fell from a seasonally adjusted rate of nearly 2.3 million in early 2006 to a low of less than 500,000 units in December 2007. Housing starts remained below one million homes per year even five years after the start of the recession. 17 There is some debate as to what extent the housing market collapse was a cause of the recession and to what extent it was a result of the recession, but what is clear is that the magnitude of the collapse was unmatched.

The property tax is assessed on the value of residential (i.e., personal real estate), commercial, business, and farm real property, and in some states, personal property (e.g., automobiles). Residential real property accounts for approximately 60 percent of taxable assessments and is the largest component of the tax base by a significant margin; commercial, industrial, and farm property account for around 30 percent, and personal property accounts for less than 10 percent.18 It is difficult to square two seemingly inconsistent results: the property tax fared much better than other state and local taxes during the GR even though the property tax base collapsed.

Several papers have noted that local jurisdictions seem to have the ability to raise property tax rates to offset property value declines. Lutz, Molloy, and Shan (2011) find assessed values lag market values in many states, which help to support property tax revenues when house prices fall. They do find that, in aggregate, property tax millage rates rise when property values decline. These results are born out in analyses of specific states, including Georgia (Alm, Buschman, and Sjoquist 2011), New York (Chakrabarti, Livingston, and Roy 2014), and Florida (Ihlanfeldt 2011).

We have looked at the hypothesis that there is an asymmetric response in tax rates to changing assessed values. The argument here is that when property values are increasing the tax rate falls but not so much that revenues decline. In contrast, when property values are falling, millage rates can be increased to more than offset the decline in values. To test this hypothesis, we have collected data on assessed valuations and property tax rates for several years for all school districts in five states: Illinois (2008–11), Washington (2009–12), Virginia (2006–11), Texas (2009–3), and Ohio (2008–12).19 In all cases, we took the data from peak to trough in per-student assessed property values. Note that the peak varies across states, a reflection of how frequently properties are assessed. In all cases we have real assessed values per student and property tax millage rates at the school district level. We use these data to estimate the econometric model,
%ΔMRi=α0+α1Di+α2%ΔPVPCi·Di+α3%ΔPVPCi·(1-Di)+ɛi,
(1)
where the dependent variable %ΔMRi is the percentage change from peak to trough in district i’s property tax millage rate. We define %ΔPVPCi as the percentage change from peak to trough in property values per capita in a school district, and we define Di as a dummy variable that has a value of 1 if %ΔPVPCi is positive (i.e., D is 1 in those districts where the per student property values rose). Because all the variables of interest are in percent changes, the coefficients on α2 and α3 are elasticities. The elasticity of the tax rate with respect to tax base changes is then α2 in districts where the tax base is rising and α3 in districts where the tax base is falling. We are interested in three hypotheses. First, is α2 equal to α3? If so, then local government response to changing property base is symmetric. Second, is α2 or α3 equal to 0? If so, then the millage rate is unaffected by changes in the tax base. In this case property tax collections would fall at the same rate as the tax base; government would offset none of the change in the property tax base by changing the millage rate. Third, is α2 or α3 equal to −1? If so, then changes in the millage rate completely offset changes in the tax base. In this case property tax collections would remain constant when the base changes.

Table 2 presents our ordinary least squares (OLS) estimates of equation 1 for five states. In the final column, we report estimates from a model that pools these five states and adds a state dummy to the model. We can reject the hypothesis that government response is symmetric in all models at the 5 percent level. In all six models, we cannot reject the null that α2 = 0, suggesting that tax rates do not adjust down when property values increase. In contrast, school districts are much more likely to raise the tax rate when the tax base falls than they are to lower the tax rate when property values rise. We can reject the null that α3 = 0 in all cases. We cannot reject the hypothesis that changes in the property tax rate fully offset changes in the tax base for Illinois, Virginia, Ohio, and the pooled sample. For Washington (Texas), our estimate of α3 is statistically smaller (larger) than −1.

Table 2.
Ordinary Least Squares Estimates, Percentage Change in District Property Tax Millage Rates, Peak to Trough as a Function of the Change in Property Values Per Capita
State and Fiscal Years Included
IL 2008 to 2011WA 2009 to 2012VA 2006 to 2011TX 2009 to 2013OH 2008 to 2012Pooled Sample
1) Di −0.017 −0.073 0.031 −0.000 −0.019 0.013 
 (0.013) (0.0437) (0.090) (0.007) (0.265) (0.014) 
2) %ΔPVPCi • Di −0.034 −0.308 0.111 0.004 −0.181 −0.020 
 (0.042) (0.210) (0.436) (0.041) (0.159) (0.072) 
3) %ΔPVPCi (1 − Di−1.190 −1.536 −1.056 −0.097 −0.927 −0.959 
 (0.130) (0.106) (0.166) (0.025) (0.086) (0.036) 
Constant 0.048 0.143 −0.126 0.020 0.058 0.051 
 (0.010) (0.023) (0.036) (0.003) (0.015) (0.013) 
p-value on test α2 = α3 <0.0001 <0.0001 0.014 0.037 <0.0001 <0.0001 
N 552 279 134 1021 612 2,598 
R2 0.335 0.601 0.267 0.018 0.265 0.410 
Sample means (unweighted)       
% Δ tax rates 4.32 23.6 −0.24 3.02 9.60 6.8 
% ΔPVPCi 8.50 −4.14 −13.4 −3.39 −3.94 −1.6 
I (% ΔPVPCi > 0) 67.2 36.6 12.7 47.2 34.9 45.7 
State and Fiscal Years Included
IL 2008 to 2011WA 2009 to 2012VA 2006 to 2011TX 2009 to 2013OH 2008 to 2012Pooled Sample
1) Di −0.017 −0.073 0.031 −0.000 −0.019 0.013 
 (0.013) (0.0437) (0.090) (0.007) (0.265) (0.014) 
2) %ΔPVPCi • Di −0.034 −0.308 0.111 0.004 −0.181 −0.020 
 (0.042) (0.210) (0.436) (0.041) (0.159) (0.072) 
3) %ΔPVPCi (1 − Di−1.190 −1.536 −1.056 −0.097 −0.927 −0.959 
 (0.130) (0.106) (0.166) (0.025) (0.086) (0.036) 
Constant 0.048 0.143 −0.126 0.020 0.058 0.051 
 (0.010) (0.023) (0.036) (0.003) (0.015) (0.013) 
p-value on test α2 = α3 <0.0001 <0.0001 0.014 0.037 <0.0001 <0.0001 
N 552 279 134 1021 612 2,598 
R2 0.335 0.601 0.267 0.018 0.265 0.410 
Sample means (unweighted)       
% Δ tax rates 4.32 23.6 −0.24 3.02 9.60 6.8 
% ΔPVPCi 8.50 −4.14 −13.4 −3.39 −3.94 −1.6 
I (% ΔPVPCi > 0) 67.2 36.6 12.7 47.2 34.9 45.7 

Notes: Regressions are weighted by size of the district in the earlier year. The regression in the final column also has state effects. IL = Illinois; WA = Washington; VA = Virginia; TX = Texas; OH = Ohio.

In all, our results strongly suggest that school districts were able to offset a declining tax base during the recession by raising the tax rate. This result has important policy implications. All taxes are unpopular but the property tax is often seen as one of the most unpopular of all.20 It is highly visible because taxpayers typically pay it directly. The tax is particularly unpopular among the elderly who often face significant tax bills but have relatively modest incomes. The tax base is typically distributed across local governments in very uneven ways, which contributes to extreme fiscal disparities across jurisdictions. One advantage of the property tax, however, is that it has proven to be a stable source of revenue. Property tax revenues in the past have been relatively insensitive to the business cycles. But until the GR, virtually all of the evidence on the stability of the property tax came from episodes when the real estate market was fairly stable despite ups and downs of the economy as a whole. The experience during the GR tells us that the stability of the property tax is a more general result than we might have imagined. Property tax revenues continued to rise even during one of the greatest upheavals in the real estate market.

As we noted earlier, states have assumed a larger role in education finance over the last forty years and rely on more volatile forms of funding. This shift toward state funding may have an unintended side effect. It could have made public education funding much more sensitive to the effects of the GR. We will present some further evidence on this point in section 3.

Employment in Public K–12 Education

We now turn to the effect of the GR on employment in K–12 education. Our primary data source here is the monthly U.S. Bureau of Labor Statistics Current Employment Statistics, a monthly survey of roughly 550,000 worksites that is available back to 1939. Those data allow us to track all public sector school personnel but do not allow us to consider the impact of the recession on jobs for teachers separate from other educational employment, such as administrators, guidance counselors, and librarians, who may not be as actively involved in the classroom environment.

Figure 5 presents an index of full-time-equivalent employment in public K–12 education, the private sector, state government, and local government outside of K–12 education for the thirty-six months before and sixty months after the start of the GR. Each series is scaled so that it equals 100 at the start of the recession. Two points are of particular interest in the figure. First, figure 5 shows that employment in education followed a very different time path than employment in the private sector. Private sector employment fell sharply at the start of the recession; two years after the start of the recession private employment was 7 percent lower than at the start. In data not shown in this graph, private employment returned to its pre-recession level by March 2014. Second, the recession took a substantial toll on public education. Jobs for school employees increased slightly or were flat during the first two years of the recession but then fell dramatically. Employment in public schools had not returned to pre-recession levels more than five years after the start of the GR. In total, employment in public schools fell by 294,700 from the start of the recession until January 2013. This represents a 3.7 percent decrease in employment. Employment in K–12 schools increased slightly in calendar year 2013, adding back only 10,000 jobs. From fall 2007 through fall 2013, public school enrollment rose by 1.6 percent21 and, therefore, the drop in public school employment meant the ratio of employees (largely teachers) to students fell by 5.1 percent over this period. Pupil–teacher ratios fell from 17.4 to 16.3 between the 1989–90 and 2003–04 school year, which was a 4.5 percent decline. The GR thus wiped out thirteen years of decline in the pupil–teacher ratio in just three years.22

Figure 5.

Employment Index for Four Sectors over the Great Recession, Bureau of Labor Statistics Current Employment Statistic (Start of the Recession = 100)

Figure 5.

Employment Index for Four Sectors over the Great Recession, Bureau of Labor Statistics Current Employment Statistic (Start of the Recession = 100)

A second data source allows us to take an initial look at the distribution of lost jobs within public schools. Our analysis draws on data from the Common Core of Data, which is an annual census of public schools and school districts. Data on employment in broad job categories for all public schools are available in the U.S. Department of Education's Digest of Education Statistics.23

In figure 6 we present an index of fall employment from 1997–98 through 2013–14 school years for four broad groups of employees in public schools: teachers, teacher aides, support staff, and other employees (which includes district administrators, principals, librarians, and guidance counselors). We set each index equal to 100 in the 2007–08 school year. As figure 6 shows, the number of teachers rose at the slowest rate among the four groups in the education sector between 1997–98 and 2008–09—teachers and support staff increased 14 percent, other employees 21 percent, and aides 22 percent over that period. Employment peaked in 2008–09, and over the next three school years the number of teachers fell 3.7 percent compared with a decrease of just 1.4 percent for support staff. Teachers represented 51 percent of employment in 2008 but were responsible for 63 percent of job loss over the first three years of the GR.

Figure 6.

Teachers, Aides, Support Staff, and Other Education Employees, 1997—98 to 2013—14 School Years (Index = 100 in 2007—08)

Figure 6.

Teachers, Aides, Support Staff, and Other Education Employees, 1997—98 to 2013—14 School Years (Index = 100 in 2007—08)

Figure 7 compares the effects of the most recent recession on public school employment to the effects of three previous recessions. In this figure, we present an index of K–12 employment for the four recessions and scale each time series so that employment equals 100 at the start of each recession. The horizontal axis measures months since the start of each recession. Figure 7 shows that the impact of the GR on teachers and other personnel was unparalleled. In the 1990 and 2001 recessions, public school employment continued to rise steadily despite the economic downturn. In the 1981 recession, which was much more severe than the 1990 and 2001 recessions, public school employment fell for two years but then recovered fairly quickly. Five years after the 1981 recession began, public education employment was about 3 percent higher than at the start of the recession. But as we have shown, employment in K–12 education remained 5 percent below the December 2007 level sixty months after the start of the GR.

Figure 7.

Employment Index for K—12 Education over the Last Four Recessions, Bureau of Labor Statistics Current Employment Statistics (Start of the Recession = 100)

Figure 7.

Employment Index for K—12 Education over the Last Four Recessions, Bureau of Labor Statistics Current Employment Statistics (Start of the Recession = 100)

3.  Which Schools Were Impacted the Most by the Great Recession?

We now shift gears and look at the effect of the GR on district spending for K–12 education. In this section, we examine how the structure of school finance in a state or school district affects the impact of the recession on schools.

To address this question, we have developed a balanced panel of school district data. We describe the construction of our dataset in Appendix A. Specifically, we match data from the Common Core to the financial data contained in the National Center for Education Statistics (NCES) Local Education Agency (School District) Finance Survey F-33 files. Our sample consists of annual observations for 9,692 regular school districts for the 1994–95 school year through the 2013–14 school year.24 Because of missing data, the final data set excludes many districts, mostly those with smaller enrollments.25 As a consequence, although the dataset contains only 71 percent of all regular districts, those districts account for 88 percent of all public school students.

We initially focus on one question: Which schools were affected most severely by the GR? The results from section 2 suggest that, everything else equal, school districts that relied heavily on state funding were more vulnerable than districts that relied on support from local taxes. As we showed, states generate most of their revenues from income and sales taxes, and both of these taxes fell sharply during the recession. In contrast, local governments rely primarily on property taxes, and property tax revenues were fairly stable during the GR.

We use our balanced panel of school districts to shed light on this issue. The dependent variable in this econometric work is the percent change in per student spending for the 2006–07 and 2010–11 school years. We would like to use data on income in this analysis, but income data are not available at the school district level. We therefore use the percentage change in county-level per capita income from 2007 to 2011 constructed from the Bureau of Economic Analysis's Local Area and Personal Income and Employment Regional Data.26 We also merge this with the change in the county unemployment rate over the 2007–11 period taken from the Bureau of Labor Statistics Local Area Unemployment Statistics.27 In column 1 of table 3, we regress the outcome of interest on these two variables plus the fraction of district revenues that came from state sources. In this column, we report OLS standard errors in square brackets and in parentheses are standard errors that allow for within-state correlation in the residuals in parentheses.

Table 3.
Ordinary Least Squares Estimates of Equation Explaining % Change in Current Expenditures per Pupil from 2007—08 to 2010—11
All Districts in Sample (9,619 observations)Districts in a CBSA (3,692 observations)
Covariates(1)(2)(3)(4)(5)(6)
% Δ County per capita income, 2007—11 0.195 0.040 0.069 0.213 0.027 0.122 
 (0.068) (0.030) (0.040) (0.099) (0.050) (0.079) 
 [0.014]   [0.025]   
Δ County unemployment rate, 2007—11 −1.267 0.546 0.249 −1.076 0.759 0.379 
 (0.313) (0.216) (0.225) (0.301) (0.234) (0.317) 
 [0.060]   [0.106]   
% district revenues from state sources, 2007 −0.090 −0.026 −0.017 −0.085 −0.024 −0.019 
 (0.027) (0.036) (0.036) (0.027) (0.029) (0.029) 
 [0.005]   [0.009]   
% Δ CBSA housing price index, 2007—11    0.033 −0.011 0.039 
    (0.045) (0.037) (0.065) 
    [0.011]   
% Δ State per capita income, 2007—11   0.048   −0.018 
   (0.308)   (0.328) 
Δ State unemployment rate, 07—11   −2.637   −3.106 
   (0.792)   (0.818) 
Share of state K—12 revenues from state sources 2007   −0.197   −0.217 
   (0.095)   (0.079) 
% Δ State housing price index, 2007—11   −0.029   −0.114 
   (0.074)   (0.119) 
Include state effects? No Yes No No Yes No 
R2 0.1339 0.4701 0.2067 0.1607 0.4857 0.2432 
All Districts in Sample (9,619 observations)Districts in a CBSA (3,692 observations)
Covariates(1)(2)(3)(4)(5)(6)
% Δ County per capita income, 2007—11 0.195 0.040 0.069 0.213 0.027 0.122 
 (0.068) (0.030) (0.040) (0.099) (0.050) (0.079) 
 [0.014]   [0.025]   
Δ County unemployment rate, 2007—11 −1.267 0.546 0.249 −1.076 0.759 0.379 
 (0.313) (0.216) (0.225) (0.301) (0.234) (0.317) 
 [0.060]   [0.106]   
% district revenues from state sources, 2007 −0.090 −0.026 −0.017 −0.085 −0.024 −0.019 
 (0.027) (0.036) (0.036) (0.027) (0.029) (0.029) 
 [0.005]   [0.009]   
% Δ CBSA housing price index, 2007—11    0.033 −0.011 0.039 
    (0.045) (0.037) (0.065) 
    [0.011]   
% Δ State per capita income, 2007—11   0.048   −0.018 
   (0.308)   (0.328) 
Δ State unemployment rate, 07—11   −2.637   −3.106 
   (0.792)   (0.818) 
Share of state K—12 revenues from state sources 2007   −0.197   −0.217 
   (0.095)   (0.079) 
% Δ State housing price index, 2007—11   −0.029   −0.114 
   (0.074)   (0.119) 
Include state effects? No Yes No No Yes No 
R2 0.1339 0.4701 0.2067 0.1607 0.4857 0.2432 

Notes: There are 9,616 observations in each regression. The numbers in parentheses are standard errors allowing for arbitrary correlation in errors across districts within a state. The numbers in square brackets are ordinary least squares standard errors. CBSA = Core Based Statistical Area.

Looking at the parameter estimates and the OLS standard errors, we find what we think most people would predict: Expenditures per pupil fell sharply in school districts where the unemployment rate rose or per capita income fell. The results also suggest that districts with greater support from the state in 2006–07 experienced significantly lower growth over the next five years.28 It is not clear, however, if the estimates in the first column of table 3 capture district-level characteristics or state-level characteristics. One piece of evidence on this question is that when we allow an arbitrary correlation in errors within a state, the standard errors increase by a factor of five, suggesting there is some shock that is common to districts within states. In the second column, we add state effects to the model and cluster the standard errors at the state level. Note that the coefficients on change in income and state share are now no longer statistically significant, and the coefficient on the unemployment rate is actually the wrong sign. This suggests that the results in column 1 capture events at the state level rather than what is happening at the district level. In the third column we get some sense of the variables that are driving this result. We drop the state effects and add in four variables measured at the state level: (1) the percentage change in real per capita income from 2007–11,29 (2) the change in the state unemployment rate,30 (3) the share of K–12 revenues provided by the state, and (4) the change in house prices from June 2007 to June 2011, as measured by the state housing price index from Freddie Mac.31 In this model, the only two statistically significant variables are the change in the state unemployment rate and the state share of education revenues for all districts in the state. Nationwide, the unemployment rate rose from 4.6 in June 2007 to 9.1 percent in June 2011. From the results in column 3, a change this large at the state level is estimated to reduce spending per student by (0.045) · (−2.64) = −0.119 or almost 12 percent. It is not uncommon to see a 20-percentage point difference in the state share in K–12 revenues across states, and a change this big is estimated to reduce spending during the recession by 4 percent.

The results in the first three columns of table 3 can be criticized because we do not include a measure of changes in property value at the local level. Data to construct such a measure are not available for all districts. However, the Freddie Mac housing price index is calculated at the level of the Core Based Statistical Area (CBSA).32 In the final three columns of table 3 we reduced the sample to the 3,692 districts that are located in CBSAs, and so we can add a local housing price index as a covariate when we use this restricted sample. The basic results in this case are qualitatively the same as the full sample. We find in column 4 that there are large effects of local economic conditions on spending that seem to be proxying for state-level variables, such as state effects (column 5) or state-level economic variables (column 6).

Table 4 summarizes our econometric estimates of some of the determinants of the impact of the recession on the district-level pupil–teacher ratio. The dependent variable in those three models is the percent change in the district pupil–teacher ratio between 2007 and 2011. The models in columns 1 to 3 mirror the structure of the estimates in table 3 where numbers in brackets represent regular OLS standard errors and the numbers in parentheses represent the standard errors clustered at the state level. Because we do not have school district level measures of unemployment and income, we use county-level aggregates for these variables.

Table 4.
Ordinary Least Squares Estimates of Equation Explaining % Change in Pupil—Teacher Ratio from 2007—08 to 2011—12
All Districts in Sample (9,619 observations)Districts in a CBSA (3,692 observations)
Covariates(1)(2)(3)(4)(5)(6)
% Δ County per capita income, 2007—11 −0.071 −0.005 −0.031 −0.143 0.046 −0.029 
 (0.075) (0.029) (0.036) (0.158) (0.046) (0.097) 
 [0.016]   [0.032]   
Δ County unemployment rate, 2007—11 0.639 −0.288 0.419 1.229 −0.351 0.956 
 (0.717) (0.165) (0.219) (0.613) (0.390) (0.583) 
 [0.083]   [0.151]   
% district revenues from state sources, 2007 0.107 0.013 0.015 0.083 −0.013 −0.017 
 (0.050) (0.017) (0.023) (0.054) (0.022) (0.031) 
 [0.007]   [0.012]   
% Δ CBSA housing price index, 2007—11    0.069 −0.012 −0.037 
    (0.112) (0.049) (0.087) 
    [0.013]   
% Δ State per capita income, 2007—11   −0.212   −0.411 
   (0.438)   (0.390) 
Δ State unemployment rate, 2007—11   1.491   1.050 
   (1.250)   (1.580) 
Share of state K—12 revenues from state sources, 2007   0.275   0.286 
   (0.118)   (0.137) 
% Δ State housing price index, 2007—11   0.203   0.328 
   (0.151)   (0.122) 
Include state effects? No Yes No No Yes No 
R2 0.0413 0.4455 0.0925 0.0411 0.4758 0.0813 
All Districts in Sample (9,619 observations)Districts in a CBSA (3,692 observations)
Covariates(1)(2)(3)(4)(5)(6)
% Δ County per capita income, 2007—11 −0.071 −0.005 −0.031 −0.143 0.046 −0.029 
 (0.075) (0.029) (0.036) (0.158) (0.046) (0.097) 
 [0.016]   [0.032]   
Δ County unemployment rate, 2007—11 0.639 −0.288 0.419 1.229 −0.351 0.956 
 (0.717) (0.165) (0.219) (0.613) (0.390) (0.583) 
 [0.083]   [0.151]   
% district revenues from state sources, 2007 0.107 0.013 0.015 0.083 −0.013 −0.017 
 (0.050) (0.017) (0.023) (0.054) (0.022) (0.031) 
 [0.007]   [0.012]   
% Δ CBSA housing price index, 2007—11    0.069 −0.012 −0.037 
    (0.112) (0.049) (0.087) 
    [0.013]   
% Δ State per capita income, 2007—11   −0.212   −0.411 
   (0.438)   (0.390) 
Δ State unemployment rate, 2007—11   1.491   1.050 
   (1.250)   (1.580) 
Share of state K—12 revenues from state sources, 2007   0.275   0.286 
   (0.118)   (0.137) 
% Δ State housing price index, 2007—11   0.203   0.328 
   (0.151)   (0.122) 
Include state effects? No Yes No No Yes No 
R2 0.0413 0.4455 0.0925 0.0411 0.4758 0.0813 

Notes: There are 9,755 observations in each regression. The numbers in parentheses are standard errors allowing for arbitrary correlation in errors across districts within a state. The numbers in square brackets are ordinary least squares standard errors. CBSA = Core Based Statistical Area.

In column 1, where we add in covariates at the county level, we see that rising income and falling unemployment rates reduce the pupil–teacher ratio. These estimates are statistically significant at conventional levels when regular OLS standard errors are used but the standard errors increase appreciably when we cluster at the state level. With either standard error, the fraction of revenues from the state is positive and statistically significant, suggesting that districts heavily dependent on the state saw an increase in the pupil/teacher level. In column 2, none of these results remains statistically significant when we add state fixed effects. When we take out the state effects and add in state variables in column 3, the only statistically significant variable is the fraction of K–12 revenues from the state across all districts. The results are essentially unchanged when we reduce the sample to districts in metro areas. However, the percentage change in the state-level housing price index is positive and statistically significant.

4.  The Impact of the Recession on School Spending Inequality

In the previous sections we focused largely on how the GR altered average spending outcomes. Here we turn to the impact of the recession on inequality in education spending. Inequality in income and wealth rose throughout much of the last half-century. Inequality in education spending, however, fell during much of that period (Murray, Evans, and Schwab 1998; Corcoran and Evans 2015). Here we ask whether this trend continued during the GR.

We consider inequality in current expenditures per pupil from unified districts over a long period. For data prior to 1993, we use measures of inequality reported by Evans and Schwab and their various coauthors, whereas for 1993 and on, we rely on the panel data from the previous section. In all, we have data for ten-year intervals from 1972 until 1992, 1990, and annual data from 1993.

We look at four measures of inequality.33 The Gini coefficient ranges from 0 (perfect equality) to 1 (perfect inequality) and is based on the Lorenz curve. The 95/5 ratio is the ratio of spending for the student at the 95th percentile in district spending divided by the student at the 5th percentile in spending. The coefficient of variation equals the ratio of the standard deviation to the mean of school spending. The final measure is the Theil index, which has the convenient property that it can be decomposed into between- and within-state measures of inequality.

The time series for the four national estimates of between-district inequality are presented in table 5. In the first row, we report the baseline levels of the inequality measures. After that, for ease of exposition, we report indexes of the measures with the baseline year (1972) equal to 1.00. Looking at the entire 42-year period between 1972 and 2013, we see very little change overall in the inequality of school spending. This is true regardless of which measure of inequality we use. This overall trend, however, masks enormous differences within this time period. There are two distinct sub-periods. From 1972 to 2000, inequality fell sharply. Depending on which measure we use, inequality in school spending fell between 21 and 39 percent during that period. The United States did make significant progress in its effort to reduce the disparities between rich and poor school districts.

Table 5.
Inequality in District-Level per Pupil Current Expenditures on K—12 Education, 1972—2013
Between- and Within-State Theil Index
YearGini (×100)95-to-5 ratioCoefficient of Variation (×100)Theil Index (×1000)Within StatesBetween States% Within% Between
 Actual values at baseline   
1972 16.2 2.73 30.57 43.1 14.0 29.2   
 Index values with 1972 = 1.00   
1972 1.00 1.00 1.00 1.00 0.32 0.68 32.4 67.6 
1982 0.85 0.81 0.84 0.71 0.32 0.39 45.0 55.0 
1990 0.97 0.93 0.98 0.95 0.30 0.65 31.3 68.7 
1992 0.93 0.90 0.96 0.90 0.28 0.63 30.4 69.6 
1993 0.90 0.88 0.93 0.84 0.26 0.58 30.7 69.3 
1994 0.86 0.85 0.91 0.80 0.23 0.57 28.8 71.2 
1995 0.85 0.83 0.90 0.77 0.22 0.55 28.2 71.8 
1996 0.82 0.81 0.87 0.73 0.21 0.52 28.4 71.6 
1997 0.79 0.79 0.84 0.68 0.19 0.48 28.6 71.4 
1998 0.75 0.77 0.80 0.62 0.19 0.43 30.5 69.5 
1999 0.75 0.77 0.80 0.61 0.19 0.43 30.4 69.6 
2000 0.74 0.76 0.79 0.61 0.19 0.42 30.6 69.4 
2001 0.74 0.76 0.79 0.61 0.19 0.42 31.4 68.6 
2002 0.76 0.77 0.81 0.64 0.20 0.44 31.1 68.9 
2003 0.78 0.78 0.83 0.67 0.20 0.46 30.4 69.6 
2004 0.81 0.79 0.87 0.73 0.22 0.51 29.8 70.2 
2005 0.83 0.80 0.90 0.77 0.22 0.55 28.7 71.3 
2006 0.83 0.81 0.92 0.80 0.23 0.57 29.1 70.9 
2007 0.84 0.82 0.93 0.81 0.23 0.58 28.6 71.4 
2008 0.85 0.83 0.94 0.84 0.24 0.59 29.1 70.9 
2009 0.84 0.83 0.91 0.79 0.22 0.57 28.0 72.0 
2010 0.86 0.85 0.94 0.83 0.22 0.61 26.8 73.2 
2011 0.89 0.88 0.96 0.87 0.23 0.65 25.7 74.3 
2012 0.93 0.90 1.02 0.97 0.23 0.74 23.6 76.4 
2013 0.94 0.91 1.04 1.01 0.23 0.78 22.6 77.4 
Between- and Within-State Theil Index
YearGini (×100)95-to-5 ratioCoefficient of Variation (×100)Theil Index (×1000)Within StatesBetween States% Within% Between
 Actual values at baseline   
1972 16.2 2.73 30.57 43.1 14.0 29.2   
 Index values with 1972 = 1.00   
1972 1.00 1.00 1.00 1.00 0.32 0.68 32.4 67.6 
1982 0.85 0.81 0.84 0.71 0.32 0.39 45.0 55.0 
1990 0.97 0.93 0.98 0.95 0.30 0.65 31.3 68.7 
1992 0.93 0.90 0.96 0.90 0.28 0.63 30.4 69.6 
1993 0.90 0.88 0.93 0.84 0.26 0.58 30.7 69.3 
1994 0.86 0.85 0.91 0.80 0.23 0.57 28.8 71.2 
1995 0.85 0.83 0.90 0.77 0.22 0.55 28.2 71.8 
1996 0.82 0.81 0.87 0.73 0.21 0.52 28.4 71.6 
1997 0.79 0.79 0.84 0.68 0.19 0.48 28.6 71.4 
1998 0.75 0.77 0.80 0.62 0.19 0.43 30.5 69.5 
1999 0.75 0.77 0.80 0.61 0.19 0.43 30.4 69.6 
2000 0.74 0.76 0.79 0.61 0.19 0.42 30.6 69.4 
2001 0.74 0.76 0.79 0.61 0.19 0.42 31.4 68.6 
2002 0.76 0.77 0.81 0.64 0.20 0.44 31.1 68.9 
2003 0.78 0.78 0.83 0.67 0.20 0.46 30.4 69.6 
2004 0.81 0.79 0.87 0.73 0.22 0.51 29.8 70.2 
2005 0.83 0.80 0.90 0.77 0.22 0.55 28.7 71.3 
2006 0.83 0.81 0.92 0.80 0.23 0.57 29.1 70.9 
2007 0.84 0.82 0.93 0.81 0.23 0.58 28.6 71.4 
2008 0.85 0.83 0.94 0.84 0.24 0.59 29.1 70.9 
2009 0.84 0.83 0.91 0.79 0.22 0.57 28.0 72.0 
2010 0.86 0.85 0.94 0.83 0.22 0.61 26.8 73.2 
2011 0.89 0.88 0.96 0.87 0.23 0.65 25.7 74.3 
2012 0.93 0.90 1.02 0.97 0.23 0.74 23.6 76.4 
2013 0.94 0.91 1.04 1.01 0.23 0.78 22.6 77.4 

Notes: Data for 1972—92 are from Corcoran and Evans (2009). All other calculations are by authors.

But the trend changed dramatically in 2000. All four of our measures reversed course and rose steadily from 2000 to 2013. The increase in inequality during this fourteen-year period undid virtually all of the progress that was made during the preceding twenty-nine years. By 2013, the coefficient of variation and the Thiel index were essentially at their 1972 values. Although finance reform was able to reduce inequality, the success was fleeting.

The last four columns of table 5 exploit the unique properties of the Theil index of inequality and look at the national, between-state, and within-state measures of current expenditures per pupil. For the within- and between-state measures, we divide by the total Theil index in 1972 so these two measures sum to the indexed value. Both measures fell sharply from 1972 to 2000. Within-state inequality fell by 41 percent and between-state inequality fell by 38 percent in this period. There is a fair amount of evidence that the decline in inequality in school spending was at least in part the result of successful court-ordered state efforts to reduce inequality. Murray, Evans, and Schwab (1998), for example, found that court-ordered education finance reform did significantly decrease within-state inequality in spending.

Although rising between- and within-state inequality both contributed to the overall increase in inequality in education spending starting in 2000, the main factor here is growing differences between the states in school spending. Between-state inequality rose by 86 percent between 2000 and 2013; within-state inequality in spending rose by just 21 percent during the period.

It is not clear that the GR played a major role in the increase in inequality in school spending. All of our inequality measures rose at nearly the same rate between 2000 and 2008 and between 2008 and 2013. The data suggest that a combination of factors other than the effect of the GR led to a continual rise in inequality in education expenditures over the fourteen-year period. Almost certainly the increase in income and wealth inequality in the United States played a key role.

The results from table 2 suggest an additional answer. When housing prices are increasing, the elasticity of response on the millage rate is negative but much less than 1 in absolute value. This suggests that as property values increase, tax revenues increase as well. Hence, in an era of rising property values, we would expect faster growth in per student spending in areas with faster growth in per student assessments. If the high-spending states in the early 2000s were also those states with the fastest growth in property value, then this would help explain why spending in these states grew so much. We do not have per student assessments for all states over the past fourteen years but we have a reasonable proxy—the change in housing prices over that period. In figure 8, we graph the percentage change in the Freddie Mac state-level house price index between 2000 and 2011 versus the real state-level per pupil current expenditures in 2000. Note that this graph has a steep positive slope, implying high-spending states in 2000 had the greatest increase in house prices over the next eleven years. If the elasticity of tax rates with respect to assessed valuations is less than 1 in absolute value when property values are increasing—a result we found for five states in table 2—this would help to explain the growth in the between-state inequality in expenditures.

Figure 8.

Percentage Change in State Housing Price Index 2000 to 2011 vs. Current Expenditures per Pupil in 2000

Figure 8.

Percentage Change in State Housing Price Index 2000 to 2011 vs. Current Expenditures per Pupil in 2000

5.  The Federal Government's Response to the Great Recession

The Stimulus Bill and Education Finance

On 17 February 2009, President Obama signed into law the American Recovery and Reinvestment Act of 2009 (ARRA).34 ARRA was a key element of the government's effort to fight the recession that began in December 2007 and accelerated with the Lehman Brothers bankruptcy in September 2008. The estimated cost of the bill when it was passed was $787 billion.35 ARRA included $237 billion in tax relief for individuals and $51 billion of tax cuts for businesses. The bill provided $155.1 billion for health care (largely additional spending on Medicaid, Medicare, and subsides for private insurance for people who were laid off from their jobs), $105.3 billion for infrastructure investment, and $82.2 billion for extended unemployment benefits and other aid to low-income workers, the unemployed, and the elderly.

ARRA provided nearly $100 billion for education. The single largest component of the aid to education was the State Fiscal Stabilization Fund (SFSF), a new, one-time appropriation of $53.6 billion. Of this amount, the U.S. Department of Education (USDOE) awarded governors approximately $48.6 billion by formula in exchange for a commitment to advance essential education reforms to benefit students from early learning through postsecondary education.36 The purpose of these funds was to help stabilize state and local government budgets in order to minimize and avoid reductions in education and other essential public services. Program funds could also be used to help support the modernization and renovation of school facilities.

The ARRA specified that 61 percent of a state's allocations were based on a state's relative population of individuals aged 5 to 24 years, and 39 percent were based on its relative share of total population. States were required to use 81.8 percent of SFSF funds for the support of public elementary, secondary, and higher education, and (as applicable) early childhood education programs and services. The states had to use their allocations to help restore Fiscal Years (FY) 2009, 2010, and 2011 support for public elementary, secondary, and postsecondary education to the greater of the FY 2008 or FY 2009 level. The funds needed to restore support for elementary and secondary education had to be run through the state's education funding formulae. If any SFSF funds remained after the state had restored state support for elementary and secondary education and higher education, the state was required to award the funds to school districts on the basis of the relative Title I shares (but not subject to Title I program requirements). States were to use the remaining 18.2 percent of their SFSF funds for education, public safety, and other government services. This could include assistance for early learning, K–12 education, and support of public colleges and universities. In addition, states could use these funds for modernization, renovation, or repair of public school and public or private college facilities.

The Education Job Funds program, signed into law on 10 August 2010, was the second major federal effort to offset some of the effects of the GR on public education.37 It provided $10 billion in assistance to save or create education jobs for the 2010–11 school year. The USDOE determined the allocation for each state38 by formula on the basis of (1) its relative population of individuals who are aged five to twenty-four years, and (2) its relative total population.39 States were required to distribute these funds to school districts either through the state's primary elementary and secondary education funding formula(e) or on the basis of the districts’ relative shares of funds under Part A of Title I of the Elementary and Secondary Education Act of 1965 for the most recent fiscal year for which data were available.

There have been several efforts to evaluate the impact of SFSF and the Education Job Funds. In their studies of the New Jersey and New York experiences, Chakrabarti and Livingston (2013a, b) found that stimulus funds were an effective stopgap when state funding fell sharply at the start of the recession. They also found that when the stimulus funding ended in 2011, education spending in both states fell because state and local funding sources had not fully recovered. A 2009 analysis by the Obama administration looked at state funding for K–12 and higher education for the two previous and current school years.40 It concluded that SFSF funds restored nearly 100 percent of the 2008–09 budget gaps and a significant portion of the 2009–10 shortfalls. Based on an analysis of states’ initial and preliminary submissions of the first ARRA quarterly reports, the study argued that over 250,000 education jobs were retained or created through ARRA.41

The Distribution of Stimulus Funds

To begin an analysis of how the stimulus impacted schools, we first merged data on stimulus payments to schools from the recovery.gov database to data from the Common Core and F-33 outlined earlier. This was a difficult exercise because the primary identification used in the recovery.gov dataset was the Duns number, which is not reported in NCES datasets. To match the data, we first selected grants for K–12 education based on the Catalog of Federal Domestic Assistance (CFDA) number, then matched specific grants to recipients based on the ZIP Code of the recipient, and then visually identified whether the name of the recipient matched the name of the district (by hand in many districts). We matched grants to the school year 2008–09 F-33 data.

A similar exercise was conducted by the Institute of Education Sciences (IES) of the USDOE (Garrison-Mogren, Gutmann, and Bachman 2012). In general, our estimates of dollar amounts are slightly higher than the values in the IES report, mainly because we used a more recent version of the recovery.gov data. In aggregate, our numbers for the SFSF grants differ by less than 1 percent from the IES report and across all grants by only 2.7 percent. We do, however, have a smaller number of districts because we match our data to the F-33 instead of the Common Core. Our numbers for funds differ most for preschool grants, where we find fewer districts but record more grant dollars.

SFSF Funds and School Finance: Some Econometric Evidence

In this section, we examine the impact of SFSF funding on state funding for education. Consider a panel dataset where we have state revenues per pupil over time for a sample of districts. Let Sijt/Pupilijt be the state funding per pupil in district i, in state j in year t and let SFSFijt/Pupilijt be the corresponding values distributed through the SFSF program. We would like to estimate a simple regression of state spending per student on SFSF funding per student of the form
Sijt/Pupilijt=β0+SFSFijt/Pupilijtβ1+μij+νjt+ɛijt,
(2)
where μij is a district fixed effect, νjt is a state-specific year effect, and ɛijt is a random error. If the SFSF program is working as intended, then the coefficient β1 should be −1; that is, an additional dollar from the federal government through this program allowed states to reduce spending by exactly a dollar. In contrast, if β1 equals 0, then SFSF had no impact on state spending.42

We face two problems in estimating equation 2. First, the recovery.gov Web site records when grants were first awarded but it does not indicate when they were spent. In the case of SFSF, there is tremendous variation across states in when these funds were distributed to districts. Some states, such as California and Illinois, spent the vast majority of funds in FY 2008, whereas other states such as Virginia reserved a significant portion of funds for distribution in FY 2009 and FY 2010. Unfortunately, we do not have a comprehensive source of data that indicates when states distributed funds to the districts. Starting in FY 2009, districts reported three variables in the F-33 associated with ARRA funding. One measured new Title 1 funds, the second measured capital outlays, and the third measured all funds that went to current expenditures. This third variable would include SFSF funds, but it would also include other funds, such as support for special education. We use this third measure as a proxy for SFSF funds, but because this category represented 81 percent of noncapital, non-Title 1 funds, it is a quite sensible proxy.

Second, there is a potential endogeneity problem in the econometric estimation. If states distributed more funds to districts with lower local revenues, then the size of SFSF awards may signal something about the underlying financial health of the local area. We believe the way in which SFSF funds were distributed at the state and local levels means that this endogeneity issue can be addressed in a straightforward two-stage least-squares (2SLS) procedure.

As we noted earlier, the allocation of SFSF funds had two distinct steps. First, states received funds from the federal government. State allocations were determined by formula and states received a specific amount based on the size of the student population and the size of the state. Second, districts had to apply for SFSF funds. Because receipt of funds required districts to agree to some specific reforms and enhanced reporting, some districts chose not to participate in the SFSF program. Our data indicate that 2 percent of all school districts did not receive SFSF funds. Nonparticipating districts tended to be smaller than average (the average size of districts not receiving funds was 2,106 students, whereas the corresponding number for participating districts was 4,245), have fewer poor students (78 versus 86 percent receiving Title 1), and were in rural areas (58 versus 55 percent).43 Overall, districts not receiving SFSF funds represent only 1.1 percent of all students and 2.1 percent of all districts in our sample.

In general, states distributed SFSF money similarly to how they distributed other state funds. In a regression of the SFSF share from the state on the state share of total revenues in FY 2008, the coefficient (standard error) on this variable is 1.007 (0.013) and we cannot reject the null that this coefficient equals 1. This is not surprising given the structure of SFSF as described earlier in this section of the paper. The fact that states distributed SFSF funds in a similar way to other state revenues will later be exploited in a 2SLS procedure. In particular, once the total size of the state's SFSF grant was announced, districts had a good sense of how much they would receive from the state. Let SFSFjt be the total SFSF funds distributed by state j in year t, let θij08 be the share of state j’s funds distributed to district i in FY 2008, and Pupilsijt be the number of students from district i in year t. Our instrument for SFSFijt/Pupilijt is then
INSTijt/Pupilijt=θij08SFSFjtPupilsijt.
(3)
In year t, district i could expect to receive θij08 × 100 percent of total SFSFjt funds distributed under this program in year t; dividing this by district size turns it into a per pupil amount. In this case we will use data from FY 2006 through FY 2011, which gives us three years before SFSF and three years when all SFSF funds were supposed to be distributed. In years prior to the ARRA, SFSFjt=0 so INSTijt/Pupilijt=0. Because the instrument only makes sense for districts that received SFSF funds, we delete districts that did not participate in that program. In all, our dataset has information for 9,450 districts over six years. In our 2SLS models we weight the regressions by number of pupils and we cluster standard errors at the state level.

In the first column of panel A in table 6, we report the first-stage estimate where we regress SFSFijt/Pupilijt on INSTijt/Pupilijt plus district effects and state-specific year effects. The coefficient on the instrument is 0.83 and we can easily reject the null that the parameter equals 0. The first-stage F-test is 66.8, so there are no concerns of finite sample bias in this case. We report the 2SLS estimate of equation 3 in the second column of panel A in table 6. The estimated coefficient on SFSFijt/Pupilijt is −0.94 with a standard error of 0.256. We therefore can easily reject the null that the coefficient is zero. We cannot, however, reject the hypothesis that federal SFSF money fully offset the decrease in state funds for education, that is, that the coefficient of interest is equal to −1; the p-value for a test of the null hypothesis where the coefficient equals −1 is 0.80. The results in panel A table 6 therefore suggest that the SFSF program worked as intended; for every dollar distributed to the districts through this program, the states reduced spending by a dollar and there was no tax relief offered from these grants to local taxpayers.

Table 6.
Two-Stage Least Squares Results of Impact of Stimulus Spending on School District Outcomes, Balanced Panel of Districts, 2005—06 to 2010—11
Panel A: Impacts to Per Pupil Revenues and Employment
OLS2SLS
Dependent Variable: CovariatesSFSFijt /PupilijtSijt /PupilijtLocal rev./PupilTotal rev./PupilEmployment/Pupil
INSTijt/Pupilijt 0.831     
 (0.102)     
SFSFijt/Pupilijt  −0.936 0.103 0.239 −0.000 
  (0.256) (0.206) (0.258) (0.000) 
1st stage F 66.8     
(p-value) (<0.0001)     
R2 0.887 0.429 0.2668 0.3668 0.1851 
 Panel B: Falsification Tests—–Impacts on Student Demographics 
 OLS 2SLS 
Dependent Variable: Covariates SFSFijt /Pupilijt Fraction Black Fraction Hispanic Fraction IEP Fraction Free/Reduced Lunch 
INSTijt/Pupilijt 0.824     
 (0.105)     
SFSFijt/Pupilijt  −0.00002 −0.00000 0.00000 −0.000018 
  (0.00000) (0.00000) (0.00000) (0.00002) 
1st stage F 61.91     
(p-value) (<0.0001)     
R2  0.2448 0.4155 0.4064 0.4407 
Panel A: Impacts to Per Pupil Revenues and Employment
OLS2SLS
Dependent Variable: CovariatesSFSFijt /PupilijtSijt /PupilijtLocal rev./PupilTotal rev./PupilEmployment/Pupil
INSTijt/Pupilijt 0.831     
 (0.102)     
SFSFijt/Pupilijt  −0.936 0.103 0.239 −0.000 
  (0.256) (0.206) (0.258) (0.000) 
1st stage F 66.8     
(p-value) (<0.0001)     
R2 0.887 0.429 0.2668 0.3668 0.1851 
 Panel B: Falsification Tests—–Impacts on Student Demographics 
 OLS 2SLS 
Dependent Variable: Covariates SFSFijt /Pupilijt Fraction Black Fraction Hispanic Fraction IEP Fraction Free/Reduced Lunch 
INSTijt/Pupilijt 0.824     
 (0.105)     
SFSFijt/Pupilijt  −0.00002 −0.00000 0.00000 −0.000018 
  (0.00000) (0.00000) (0.00000) (0.00002) 
1st stage F 61.91     
(p-value) (<0.0001)     
R2  0.2448 0.4155 0.4064 0.4407 

Notes: Standard errors in parentheses allow for arbitrary correlation in errors within a state. Panel A: There are 6 observations per district for 9,450 districts for 56,700 observations. Other covariates in the model are state-specific year effects and district fixed-effects. We cluster the standard errors at the state level. Results are weighted by annual student enrollment. Panel B: As a falsification test, estimates in Panel B explore the impact of SFSF funds on district demographics. Due to a small number of districts not reporting demographic information every year, the balanced sample is slightly smaller than in Panel A. The first stage estimates are not greatly altered as a result of this. Other covariates in the model are state-specific year effects and district fixed effects. We cluster the standard errors at the state level. Results are weighted by annual student enrollment. 2SLS = two-stage least-squares; IEP = individual education plan; OLS = ordinary least squares; SFSF = State Fiscal Stabilization Fund.

We examine the impact of SFSF spending on several key variables in the final three columns of panel A table 6. Here we replace state funding per pupil as the outcome with local revenues per student (column 3), total revenues per student (column 4), or education employment per student (column 5). In 2SLS models, we find that SFSF funds are uncorrelated with these three variables. Total spending would include SFSF spending and so this result implies that federal funds were able to fully offset decreases in state and local funding.

We present estimates of a falsification analysis of our model in panel B of table 6. In this analysis we consider several school district characteristics that should not be affected by SFSF spending. These characteristics are the fraction of students who are black or Hispanic, have an individual education plan (IEP), and receive free or reduced-price lunch. Estimates of the impact SFSF on these characteristics in panel B table 6 are statistically insignificant and virtually zero. This offers support for the validity of our 2SLS analysis.

6.  Summary and Conclusions

This project sought to answer one overriding question: How did the Great Recession affect public schools? Our work has produced some interesting answers but it has generated some important further questions as well. Our results strongly suggest that the growing role of the states in public education magnified the impact of the GR. Over the last forty years, states have assumed more and more responsibility for funding public schools. This shift in the way schools are financed is in part a result of legislative and judicial efforts to reduce the wedge between resources in rich and poor schools. As previous work has shown, the increase in state funding successfully reduced between-district inequality from 1972 through 2000. But this growing reliance on the states has proved costly. States rely on taxes that are particularly sensitive to the ups and downs of the economy. The GR led to a steep decrease in state tax revenues, and consequently, state support for schools. An important question we do not address is whether it is possible to have a more redistributive state system of support that does not subject districts to these large systemic shocks. Given that states will continue to rely on income and sales taxes, these goals seem incompatible.

We also found the surprising result that the GR had a much smaller impact on local support for education. The recession began with a sharp fall in housing prices. Given that most local school districts rely heavily on the property tax, one would expect that local tax revenues would have fallen. This turned out not to be the case. As we showed in some of our state case studies, school districts were able to raise tax millage rates to compensate for any loss in property values. In four of five states we studied, we cannot reject the null that when property values were falling, the elasticity of the millage rate with respect to per capita property values was −1. The ease with which local governments can change the property tax stands in stark contrast to voter sentiment about the property tax. Our results suggest one possible reason voters dislike the property tax is that local government spending always rises. When property values per capita are increasing, the millage rate is left unchanged, indicating that total revenues per pupil increase in those situations. In contrast, when property value declines, the millage rate is increased enough to make up for any revenue shortfall. Local revenues per pupil are almost immune from reductions.

Starting in the early 2000s, the long-term secular decline in between-district inequality in per student spending reversed, and now inequality is about the same as it was in 1972. In a span of twelve years, thirty years of reductions in spending inequality were eliminated. This occurred well before the start of the GR, and hence, this trend had little to do with the recession. In fact, the available evidence suggests that GR may have reduced inequality slightly.

This all leaves open the question of why things changed so dramatically in the early 2000s. Corcoran and Evans (2010) argue that in a simple median-voter model, rising inequality due to changes in the top of the income distribution should increase public spending. With an income or wealth tax, rising inequality at the top of the distribution reduces the cost to the median voter of raising an extra dollar of revenues and hence increases spending. This is one possible mechanism, but prior to this paper, no one has documented the rising inequality in between-district spending over the past decade, let alone identified a possible mechanism.

Finally, the Great Recession led to the largest expansion of federal support for public education. Through ARRA, the federal government's share of school district revenue nearly doubled, albeit for a short period of time. The largest component of ARRA support for education came through the State Fiscal Stabilization Funds. Our results suggest that the program worked as intended—a dollar in ARRA support offset a dollar of state funding for education. Unfortunately, our results do not answer a more difficult question, which is what states would have done had they not received stimulus funds.

Notes

3. 

See www.bls.gov/news.release/empsit.t12.htm. The U.S. Bureau of Labor Statistics classifies someone as long-term unemployed if they have been unemployed for at least twenty-seven weeks.

4. 

Unless otherwise noted, years here are fiscal years. A fiscal year corresponds roughly to an academic year and so, for example, fiscal 1969 is roughly the 1968–69 academic year. Current expenditures include salaries, employee benefits, purchased professional and technical services, purchased property and other services, and supplies. It also includes gross school system expenditure for instruction, support services, and noninstructional functions. It excludes expenditure for debt service, capital outlay, and reimbursement to other governments (including other school systems).

6. 

We estimated the correlation between the data series to be −0.23. Though this is not incredibly large in magnitude, it does suggest a clear negative relationship between student spending and unemployment.

7. 

In the United States, the unofficial beginning and ending dates of national recessions have been defined by the National Bureau of Economic Research (NBER). The NBER defines a recession as “a significant decline in economic activity spread across the economy, lasting more than a few months, normally visible in real gross domestic product (GDP), real income, employment, industrial production, and wholesale-retail sales.”

9. 

This literature is captured in a number of papers, including Evans, Murray, and Schwab (1997); Murray, Evans, and Schwab (1998); Hoxby (2001); Card and Payne (2002); Figlio, Husted, and Kenny (2004); and more recently, Jackson, Johnson, and Persico (2016); LaFortune, Rothstein, and Schanzenbach (2018); and Candelaria and Shores (2019).

11. 

Information about state tax policies can be found at http://taxfoundation.org.

12. 

See Gordon (2012) for a further analysis of the impact of the Great Recession on state and local government finance.

13. 

See https://www.census.gov/programs-surveys/qtax/data/tables.html. The four-quarter moving average is constructed by averaging the real value of the current quarter with the three quarters prior to it.

14. 

We use the GDP implicit price deflator to generate real values and a four-quarter moving average because of very large within fiscal year variation in quarterly revenues.

18. 

Statistics are from Lutz, Molloy, and Shan (2011).

19. 

Data on local tax rates for Virginia can be found at https://tax.virginia.gov/local-tax-rates, and tax collections are found at https://tax.virginia.gov/assessment-sales-ratio-studies. Financial data for Washington are provided on an annual basis, and the data for 2012–13 school year are found at www.k12.wa.us/safs/PUB/FIN/1213/fs.asp. Data for Texas are available at https://tea.texas.gov/Finance_and_Grants/State_Funding/Additional_Finance_Resources/School_District_Property_Values_and_Tax_Rates/. Illinois data are available at http://webprod1.isbe.net/ilearn/ASP/index.asp. The data for Ohio for tax year 2012 are available at https://www.isbe.net/pages/illinois-state-report-card-data.aspx. We should note that these states were selected because we could easily retrieve the data.

20. 

See, for example, Alm (2013) and Cabral and Hoxby (2012).

24. 

According to the NCES, regular districts include local education agencies that operate primary and secondary schools but exclude such districts as regional education service agencies, supervisory union administrative centers, state or federally operated agencies, and independent charter schools.

25. 

Average enrollment in dropped districts was 3,277 students versus 4,232 students in sample districts.

27. 

State unemployment rates are from the Bureau of Labor Statistics Local Area Unemployment Statistics and were downloaded from the Department of Agriculture's Economic Research Service at www.ers.usda.gov/data-products/county-level-data-sets/download-data.

28. 

In our sample, 45.8 percent of the variation in the share of education revenue from the state in 2007 is within-state and across districts; the remaining 54.2 percent of the variation is across states.

29. 

Per capita income information at the state level can be downloaded from the Bureau of Economic Analysis at www.bea.gov/iTable/iTable.cfm?reqid=70&step=1&isuri=1&acrdn=5#reqid=70&step=25&isuri=1&7022=49&7023=7&7024=non-industry&7001=749&7029=49&7090=70.

30. 

State unemployment rates are from the Bureau of Labor Statistics Local Area Unemployment Statistics and were downloaded from the Department of Agriculture's Economic Research Service at http://www.ers.usda.gov/data-products/county-level-data-sets/download-data.aspx#.U8_727FvLkg.

31. 

These data are available for download at www.freddiemac.com/finance/fmhpi/.

32. 

A CBSA includes an urban center and its suburbs. CBSAs are defined by the Office of Management and Budget (OMB) and replaced the old OMB concept of a metropolitan area.

33. 

See Berne and Stiefel (1984) for a thorough discussion of the properties of measures of equity in public school funding.

36. 

The USDOE was to use $5 billion of funding to make competitive grants under the “Race to the Top” fund. These grants were designed to help states make significant improvement in student achievement. The USDOE was also authorized to spend up to $650 million to make competitive awards under the “Invest in What Works and Innovation” fund. These awards were to serve as rewards to districts and nonprofit organizations that had made significant gains in closing achievement gaps to serve as models for best practices.

37. 

This section draws heavily on USDOE (2011).

38. 

The amount of funding available to each state under the program is provided on the program Web site at www2.ed.gov/programs/educationjobsfund/index.html.

39. 

For purposes of this program, the District of Columbia and Puerto Rico are defined as states.

41. 

Rentner and Ushe (2012) tracked the use of ARRA and Education Jobs funds and the implementation of ARRA-related reforms. These six reports were based on survey responses of state and local officials charged with implementing these programs.

42. 

Thus, β1 equal to −1 is evidence of a complete flypaper effect and β1 equal to 0 is evidence of the absence of a flypaper effect.

43. 

These numbers are based on data from the FY 2008 F-33 data.

44. 

In some years, New York City Public Schools reported as thirty-three distinct geographic districts. We aggregated the components of these districts to form one aggregate measure in the dataset for New York City Public Schools.

Acknowledgments

We thank the Russell Sage Foundation for their generous support of this research. We thank Frances Kelsey for excellent research assistance.

REFERENCES

Alm
,
James
.
2013
.
A convenient truth: Property taxes and revenue stability
.
Cityscape
15
(
1
):
243
245
.
Alm
,
James
,
Robert D.
Buschman
, and
David L.
Sjoquist
.
2011
.
Rethinking local government reliance on the property tax
.
Regional Science and Urban Economics
41
(
4
):
320
331
.
Berne
,
Robert
, and
Leanna
Stiefel
.
1984
.
The measurement of equity in school finance: Conceptual, methodological, and empirical dimensions
. Vol.
17
.
Baltimore, MD
:
Johns Hopkins University Press
.
Bhalla
,
Ravi
,
Rajashri
Chakrabarti
, and
Max
Livingston
.
2017
.
A tale of two states: The recession's impact on New York and New Jersey school finances
.
Economic Policy Review
23
(
1
):
31
42
.
Cabral
,
Marika
, and
Caroline
Hoxby
.
2012
.
The hated property tax: Salience, tax rates and tax revolts.
NBER Working Paper No. 18514
.
Candelaria
,
Christopher A.
, and
Kenneth A.
Shores
.
2019
.
Court-ordered reforms in the adequacy era: Heterogeneous causal effects and sensitivity
.
Education Finance and Policy
14
(
1
):
31
60
.
Card
,
David
, and
A.
Abigail Payne
.
2002
.
School finance reform, the distribution of school spending, and the distribution of student test scores
.
Journal of Public Economics
83
(
1
):
49
82
.
Chakrabarti
,
Rajashri
, and
Max
Livingston
.
2013a
.
The long road to recovery: New York schools in the aftermath of the Great Recession
.
Available
https://www.newyorkfed.org/medialibrary/media/research/staff_reports/sr631.pdf.
Accessed 17 October 2018
.
Chakrabarti
,
Rajashri
, and
Max
Livingston
.
2013b
.
Still not out of the woods? New Jersey schools during the recession and beyond
.
Available
https://www.newyorkfed.org/medialibrary/media/research/staff_reports/sr632.pdf.
Accessed 17 October 2018
.
Chakrabarti
,
Rajashri
,
Max
Livingston
, and
Joydeep
Roy
.
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
.
Chakrabarti
,
Rajashri
,
Max
Livingston
, and
Elizabeth
Setren
.
2015
.
The Great Recession's impact on school district finances in New York state
.
Economic Policy Review
21
(
1
):
45
66
.
Chakrabarti
,
Rajashri
, and
Sarah
Sutherland
.
2013
.
Precarious slopes? The Great Recession, federal stimulus, and New Jersey schools
.
Economic Policy Review
19
(
2
):
41
65
.
Corcoran
,
Sean
, and
William N.
Evans
.
2010
.
Income inequality, the median voter, and the support for public education
.
NBER Working Paper No. 16097
.
Corcoran
,
Sean
, and
William N.
Evans
.
2015
.
Equity, adequacy and the evolving state role in education finance
. In
Handbook of research on education finance and policy
,
edited by
Helen F.
Ladd
and
Margaret
Goertz
, pp.
376
391
.
New York
:
Routledge
.
Evans
,
William N.
,
Sheila E.
Murray
, and
Robert M.
Schwab
.
1997
.
Schoolhouses, courthouses, and statehouses after Serrano
.
Journal of Policy Analysis and Management
16
(
1
):
10
31
.
Figlio
,
David N.
,
Thomas A.
Husted
, and
Lawrence W.
Kenny
.
2004
.
Political economy of the inequality in school spending
.
Journal of Urban Economics
55
(
2
):
338
349
.
Garrison-Mogren
,
Roberta
,
Babette
Gutmann
, and
Meredith
Bachman
.
2012
.
State and district receipt of Recovery Act funds: A report from charting the progress of education reform—An evaluation of the Recovery Act's role
.
Washington, DC
:
Institute of Education Sciences, National Center for Education Evaluation and Regional Assistance
.
Gordon
,
Tracy
.
2012
.
State and local budgets and the Great Recession. Available
www.brookings.edu/research/articles/2012/12/state-local-budgets-gordon.
Accessed 15 October 2018
.
Hoxby
,
Caroline M.
2001
.
All school finance equalizations are not created equal
.
Quarterly Journal of Economics
16
(
4
):
1189
1231
.
Ihlanfeldt
,
Keith R.
2011
.
How do cities and counties respond to changes in their property tax base
?
Review of Regional Studies
41
(
1
):
27
48
.
Jackson
,
C. Kirabo
,
Rucker C.
Johnson
, and
Claudia
Persico
.
2016
.
The effects of school spending on educational and economic outcomes: Evidence from school finance reforms
.
Quarterly Journal of Economics
131
(
1
):
157
218
.
LaFortune
,
Julien
,
Jesse
Rothestein
, and
Diane Whitmore
Schanzenbach
.
2018
.
School finance reform and the distribution of student achievement
.
American Economic Journal: Applied Economics
10
(
2
):
1
26
.
Lutz
,
Byron
,
Raven
Molloy
, and
Hui
Shan
.
2011
.
The housing crisis and state and local government tax revenue: Five channels
.
Regional Science and Urban Economics
41
(
4
):
306
319
.
Minorini
,
Paul
, and
Stephen
Sugarman
.
1999
.
School finance litigation in the name of educational equity: Its evolution, impact and future
. In
Equity and adequacy in education finance: Issues and perspectives
,
edited by
Helen F.
Ladd
,
Rosemary
Chalk
, and
Janet S.
Hansen
, pp.
34
71
.
Washington, DC
:
National Academy Press
.
Murray
,
Sheila
,
William N.
Evans
, and
Robert M.
Schwab
.
1998
.
Education finance reform and the distribution of education resources
.
American Economic Review
88
(
4
):
789
812
.
Rentner
,
Diane Stark
, and
Alexandra
Ushe
.
2012
.
What impact did education stimulus funds have on states and school districts?
Washington, DC
:
Center on Education Policy
.
Snyder
,
Thomas
.
1993
.
120 years of American education: A statistical portrait
.
Available
http://nces.ed.gov/pubs93/93442.pdf.
Accessed 17 October 2018
.
United States Department of Education (USDOE)
.
2011
.
Initial guidance for states on the education jobs program
.
Available
www2.ed.gov/programs/educationjobsfund/legislation.html.
Accessed 4 July 2016
.

Appendix A:  Construction of our Dataset

We have developed a balanced panel school district dataset for this analysis. The key source of data here is the Common Core of Data (CCD) from the NCES, which is available at the Local Education Agency (LEA) (district) level. The Universe Surveys in the CCD are available annually for the 1986–87 to 2013–14 academic years and include data on enrollment, staff counts, and basic school demographics recorded in the fall of the school year.

The Universe Surveys of the CCD, however, do not contain any financial information. We take data at the LEA level and merge the Universe Surveys to the Finance Survey (F-33) data from the CCD. The F-33 data provide information on revenues, expenditures, and capital outlays for districts, and is available annually for the 1991–92 to the 2012–13 school years. As we constructed the dataset for our analysis, we found the financial data in the F-33 contained some extremely large and small values. These values could be valid, but it is more likely that some districts incorrectly reported enrollments or expenditures. We therefore developed an algorithm to delete extreme values. First, we calculated the (unweighted) 95th and 5th percentile districts in total per-pupil current expenditures for each state and year. We then deleted districts with per-pupil expenditures greater than 150 percent of the 95th percentile of per-pupil revenues or less than 50 percent of the 5th percentile.

Along with eliminating extreme values, we further restrict our sample by only keeping school districts that had similar operating procedures and organization. Given that schools with different operating procedures and organizations often experience different costs (secondary schools are typically more expensive than elementary schools), we focus on districts that are unified across elementary and secondary schools. All schools in our sample are also regular districts.

The Universe Survey of the CCD contains staff counts broken down by position type. Certain positions, however, are more consistently reported than others (e.g., districts report the number of teachers but not all keep track of the number of library support staff). To account for this, we create a balanced panel by excluding districts that failed to report information for multiple years and districts that were only in operation for a limited number of years during our study period.44 We use five key variables to balance the districts in our dataset: teachers, guidance counselors, school administrators, school district administrators, and student enrollment. Teacher employment and student enrollment were the most consistently reported variables across districts. In order to stay in the balanced sample, a school district had to report these variables in at least seventeen of our twenty-year study period (school years 1993–94 to 2012–13). For the remaining staff variables—guidance counselors, school and school district administrators—districts needed to report at least twelve years.

Once we exclude these districts, we then develop a procedure to fill in the missing information of the remaining districts in the balanced sample. Missing teacher employment and student enrollment are linearly interpolated using data reported by a district in the year before and the year after the missing year of data. Given this strategy, we eliminate all districts that failed to report teacher or student information in the first or last years of our study period. Missing information for guidance counselors, school administrators, and district administrators is generated by calculating the average ratios of these employment types to teachers by district over the entire study period. We then multiply this ratio by the teacher count for the corresponding missing year to estimate the number of employees. The final dataset excludes many small districts. As a consequence, although the dataset contains only about 71 percent of all regular districts, those districts account for 88 percent of all public school students.