Clemens, Hoxie, Veuger, Intergovernmental Grants to School Districts and Educational Outcomes during the COVID-19 Pandemic (NBER w35447, )

Summary

NBER Working Paper 35447, dated July 2026, by Jeffrey Clemens, Philip G. Hoxie and Stan Veuger on pandemic aid to school districts through the Elementary and Secondary School Emergency Relief (ESSER) fund, which the abstract says totaled $189.5 billion. The authors study schools with district poverty shares near 5% using a difference-in-discontinuities design. They state that federal funds were passed on to residents through reductions in local revenue collections, including property taxes, and did not result in increased per-pupil expenditures. They find no evidence that additional funds mitigated the declines in test scores and provide suggestive evidence that political engagement increased.

Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used

Full text

                              NBER WORKING PAPER SERIES




   INTERGOVERNMENTAL GRANTS TO SCHOOL DISTRICTS AND EDUCATIONAL
              OUTCOMES DURING THE COVID-19 PANDEMIC

                                         Jeffrey Clemens
                                         Philip G. Hoxie
                                           Stan Veuger

                                       Working Paper 35447
                               http://www.nber.org/papers/w35447


                     NATIONAL BUREAU OF ECONOMIC RESEARCH
                              1050 Massachusetts Avenue
                                Cambridge, MA 02138
                                     July 2026

We are grateful to Jensen Ahokovi and Shantanu Kamat for their research assistance. We would
also like to thank Vincent Armantano, David Arnold, Julia Barezneva, Vafa Behnam, Julian Betts,
Jonathan Brownrigg, Richard Carson, Gabriel Canedo Riedel, Matthew Chingos, Julie Berry
Cullen, Gordon Dahl, Ricardo Di Cato, Maria de la Paz Ferro, Martha Fiehn, Erick Gong,
Amanda Gregg, Idaliya Grigoryeva, Angela Gu, John Juneau, Juan Hernandez Leal, Thomas Kane,
Beata Luczywek, Caitlin Knowles Myers, Paul Sommers, and Yuli Xu as well as conference and
seminar participants at Duke University’s Sanford School of Public Policy, Middlebury College,
the National Bureau of Economic Research Conference on Fiscal Dynamics of State and Local
Governments, the Society for Research on Educational Effectiveness, and UCSD for their helpful
comments. Clemens thanks the Hoover Institution at Stanford University for financial support in
his capacity as an adjunct Senior Fellow and the Institute for Humane Studies for financial support
in his capacity as a Distinguished Fellow. Hoxie acknowledges support from the National Science
Foundation Graduate Research Fellowship (2022295990) and thanks the Alfred P. Sloan
Foundation for its support of the NBER Fiscal and Economic Effects of Productivity Policies
Fellowship and the Institute for Humane Studies for financial support in his capacity as a Junior
Fellow (IHS020222). The views expressed herein are those of the authors and do not necessarily
reflect the views of the National Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.

© 2026 by Jeffrey Clemens, Philip G. Hoxie, and Stan Veuger. All rights reserved. Short sections
of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full
credit, including © notice, is given to the source.
Intergovernmental Grants to School Districts and Educational Outcomes During the COVID-19
Pandemic
Jeffrey Clemens, Philip G. Hoxie, and Stan Veuger
NBER Working Paper No. 35447
July 2026
JEL No. H71, H72, I22

                                           ABSTRACT

The federal government appropriated $189.5 billion in pandemic aid to school districts through the
Elementary and Secondary School Emergency Relief (ESSER) fund. We evaluate their impact on
schools with district poverty shares near 5%, where a threshold-driven increase in funds enables us
to implement a difference-in-discontinuities design. Federal funds were passed on to residents
through reductions in local revenue collections, including property taxes, and did not result in
increased per-pupil expenditures by school districts. We find no evidence that additional funds
mitigated the declines in test scores. We provide suggestive evidence that political engagement,
especially from parents, increased in districts that qualified for additional funds. This combination
of engagement and reduced taxation may explain the greater enrollment and faster reopening we
observe in such districts.

Jeffrey Clemens                                   Stan Veuger
University of California, San Diego               American Enterprise Institute
Department of Economics                           stan.veuger@aei.org
and NBER
jeffclemens@ucsd.edu

Philip G. Hoxie
University of California, San Diego
Department of Economics
phoxie@ucsd.edu
1. Introduction
The COVID-19 pandemic created significant challenges for the public sector in the United States
in general, and for public schools in particular. School districts faced the challenge of balancing
students’ educational achievement and the potential for disease transmission associated with
different learning modalities (in-person, virtual, or hybrid) (Gordon and Reber, 2020; Halloran et
al., 2021, 2023; Oster et al., 2021). In Figure 1, we show changes in test scores for math and
reading by school district poverty share. The patterns reflect widespread learning loss in districts
across the poverty spectrum.
       To address these and other challenges, Congress appropriated almost $1 trillion in
emergency funding in the form of intergovernmental transfers to sub-national governments, with
about $189.5 billion allocated for elementary and secondary education (Clemens and Veuger,
2021b; Office of Elementary and Secondary Education, 2023). The largest program for aiding
public schools was the Elementary and Secondary School Emergency Relief Fund (ESSER),
which received appropriations in each of the Coronavirus Aid, Relief, and Economic Security
(CARES), Coronavirus Response and Relief Supplemental Appropriations (CRRSA), and
American Rescue Plan (ARP) Acts. This paper conducts an empirical investigation into how
effective these funds were in helping schools weather the pandemic, using traditional and
pandemic-specific measures of school performance as well as budgetary outcome metrics.
       We use a difference-in-discontinuities design that exploits a threshold in the Title 1-A
funding formulas used to distribute ESSER funds to school districts. This discontinuity allows us
to identify the impacts of pandemic-era federal funds on school district outcomes local to our
policy discontinuity at the 5% policy threshold. ESSER funds were allocated to states using the
pre-existing Title 1-A funding formulas. State education agencies (SEAs) then distributed the
funds to school districts (LEAs) using these same formulas (Office of Elementary and Secondary
Education, 2022, 2023). We use a statutory jump in the Title 1-A formulas to identify a $388
per-pupil increase in ESSER funds for school districts with more than 5% of children ages 5–17
in poverty (Congressional Research Service, 2017). This cutoff is determined using pre-COVID
school district demographics, namely the 2018 vintage of the US Census Bureau’s Small Area
Income and Poverty Estimates (SAIPE). Therefore, the additional ESSER funds distributed to
districts that are just on either side of the 5% poverty threshold are plausibly exogenous to
pandemic-related outcomes. We then use the 5% poverty share discontinuity to evaluate whether


                                                 2
ESSER funds explain differences in traditional performance measures such as test scores,
enrollment, and absenteeism; pandemic outcomes like reopening speed and area cases; staffing
categories; and district budgets.
        We find that additional ESSER funds did not increase either math or reading test scores
in districts just above the 5% poverty qualification cutoff. While our point estimates are negative,
they are insufficiently precise to rule out meaningful increases. That said, ESSER funds may
have helped districts retain students and reduce absenteeism. We also find that ESSER funds
may have helped schools reopen faster, as suggested by cellphone mobility data.1 In our analysis
of staffing, we show that additional ESSER funds may have led to increases in the number of
administrators, but not in the number of teachers or overall non-teacher staff. Our set of
estimates on expenditures is generally too imprecise to draw strong conclusions, and there is no
evidence of detectable increases across broad expenditure categories as a result of qualifying for
additional funds.
        Our analysis of revenues sheds light on the primary reason why additional ESSER funds
had no detectable impact on expenditure. We find evidence of statistically and economically
large reductions in districts’ revenue from local sources in response to receipt of additional
ESSER funds. These results contradict the older literature on the “flypaper effect,” which
suggests that windfall revenues for subnational governments tend not to be offset by reductions
in own-source revenue (Gramlich, 1977; Hines and Thaler, 1995; Fisher, 1982). They are
complementary, however, to research from Gordon (2004), who finds that declines in local
revenues offset increases in federal grants over a horizon of two years following census-driven
updates to school districts’ funding allocations. We emphasize that this passthrough of
intergovernmental transfers to local taxpayers can explain the at most modest impacts we find on
students’ test scores.
        The policy bundle of decreased local revenues, accelerated reopening, and stable
expenditures can be rationalized by increased political engagement in districts that qualified for
additional ESSER funds. We provide evidence that, in these districts, local meetings likely had

1
  Additionally, we find no evidence that reopening was correlated with additional spread of COVID-19, the primary
justification for relatively slow rates of reopening. Data from UNESCO (2023) suggests that the US kept schools
closed longer than in the EU. In October of 2020, the US is marked as having “partially open” schools whereas EU
member countries’ schools were “open” with the exception of Czechia, which was “closed.” By the end of the
tracker in February of 2022, the EU would be completely open, whereas the US would be marked “partially open,”
the same status in which it was categorized in April of 2020.

                                                        3
higher attendance and parents made up a larger share of voters than in districts that did not
qualify for additional funds. As in the theory of collective budgets from Bradford and Oates
(1971a, 1971b), citizens preferences were reflected in the passthrough of a one-time lump-sum
grant.
         Our findings contribute to three literatures. First, our analysis of the ESSER program
adds to a body of work evaluating the broad set of COVID spending programs. These programs
include the Economic Impact Payments to households (Chetty et al., 2020), the Paycheck
Protection Program to support employment linkages (Autor et al., 2022a, 2022b; Hubbard and
Strain, 2020), the Municipal Lending Facility (Haughwout et al., 2021), and grants to state and
local governments to support the provision of public services, economic activity, and pandemic
mitigation efforts (Clemens, Hoxie, and Veuger, 2025; Clemens, Hoxie, Kearns, and Veuger,
2023; Clemens and Mahajan, 2025; Green and Loualiche, 2021). We add to this literature by
analyzing the effects of the ESSER program on school enrollment, reopening, student outcomes,
and district budgets.
         The second literature to which we contribute is the growing body of work analyzing
specific issues surrounding education during the COVID-19 pandemic.2 Early in the pandemic,
Gordon and Reber (2020) outlined how Congress could design an aid package to schools using
the experience of the Great Recession as a guide. As many schools switched to virtual learning,
and subsequently returned to in-person instruction, Oster et al. (2021) showed that Black and
Hispanic students returned to in-person learning later than White students. Halloran et al. (2021),
Halloran et al. (2023), and Goldhaber et al. (2023) evaluate how learning modalities and other
pandemic policies impacted test scores. Fahle et al. (2023) demonstrate that poorer districts with
more non-White students saw larger declines in test scores, but that differences between groups
of students within districts were small. Carbonari et al. (2024) find that tutoring and other
learning interventions had little impact on students’ scores in four large districts.
         There is also a growing literature on ESSER funds. Dewey et al. (2025) provide evidence
suggesting that additional ESSER funds are correlated with academic recovery across school
districts. However, ESSER funds themselves are derived from Title I funds, which are strongly
correlated with district area poverty (Gordon and Reber, 2023). To address this, Dewey et al.


2
 This literature is also related to the traditional school quality literature (Betts, 1995; Card and Krueger, 1992;
Hanushek, 1986; Jackson et al., 2016; Rivkin et al., 2005).

                                                            4
(2024) use random survey error in a district’s SAIPE estimates in 2018, the survey used to
apportion Title I funds and ESSER, as an instrument. When deploying this more plausibly causal
research design, they find no impact of ESSER funds on test score recovery in either math or
reading. Having implemented a plausibly causal design and estimated null effects, Dewey et al.
(2024) nonetheless place greater emphasis on their OLS estimates, despite clear concerns around
endogeneity and omitted-variable bias.3 The non-SAIPE error instrumental-variables estimates in
Dewey et al. (2024) use Title I funding shares as instruments for ESSER, which are highly
correlated with district poverty (Gordon and Reber, 2023). As poverty is correlated with a wide
variety of factors that influence educational outcomes, in the pandemic or otherwise, this
identifying variation is strongly correlated with other baseline differences across school districts.
         A related analysis by Goldhaber and Falken (2026) similarly uses a district’s 2018 area
poverty share as an instrument for ESSER funds in the cross section and reports meaningful
effect sizes. This instrument again raises strong concerns regarding the exclusion restriction.
Additionally, when Goldhaber and Falken (2026) include state fixed effects in their
specifications, their point estimates declines by 60 to 90 percent in magnitude, reinforcing the
aforementioned identification concerns and revealing significant sensitivity across specifications.
         An important takeaway from these analyses is that the relatively strongly positive point
estimates (i.e., of ESSER funds on achievement) in the existing literature are identified using
variation that is strongly correlated with district poverty measures and are thus exposed to
substantial endogeneity concerns. The more plausibly causal estimates presented in Dewey et al.
(2024) and in our own analysis tend towards zero.
         Although our difference-in-discontinuities results recover a local estimate, we contribute
to this literature by providing credibly causal evidence, applicable to the relatively small and
affluent districts near the 5% poverty share threshold, that incremental federal funds may have
supported faster rates of reopening and improved rates of student retention but likely did not
cause detectable improvements in test scores. Our difference-in-discontinuities design isolates
changes that occur in otherwise similar districts just above and just below the Title 1-A
qualification threshold for additional funds. By including district and time fixed effects, we net



3
  One important source of endogeneity arises from Dewey et al.’s (2024) use of spent ESSER funds instead of the
total ESSER funds allocated to a district. For example, districts with higher value uses for the funds may be more
likely to spend their allocation than other districts.

                                                          5
out baseline differences across these districts generated by the Title 1-A funds. Our estimates
suggest that at least for relatively smaller and more affluent districts, this may have been because
school boards elected to reduce property tax collections for their voters.
           We also contribute to the large literature on the flypaper effect.4 Early work on the
flypaper effect focused on the empirical finding that a fiscal transfer from an intergovernmental
grant increased sub-national government spending more than an equivalent increase in personal
income (Gramlich, 1977; Hines and Thaler, 1995; Fisher, 1982). However, later research
suggests that the flypaper effect may dissipate as governments adjust to new levels of transfers
(Gordon, 2004). There may in fact be no flypaper effect at all in some cases. Lutz (2010)
documents a roughly dollar-for-dollar decrease in local taxes in response to state education
grants in New Hampshire. More recently, Shoag, Tuttle, and Veuger (2019) suggest that the
flypaper effect may be muted by a local government’s ability to access and adjust other revenue
streams. We contribute to this literature by providing evidence that incremental federal funds
were largely passed on to local residents through reductions in local revenue collections, in
particular through reductions in local property tax collections. If districts around the 5% poverty
threshold tended to use ESSER funds to cut (or not increase) local revenue instead of increasing
instructional expenditures, this may at least partially explain why meaningful increases in
educational achievement are difficult to detect in our setting.
           This paper proceeds as follows. In Section 2, we discuss how the federal government
typically supports local districts and how ESSER funds were distributed. In Section 3, we turn to
the construction of our dataset, and in Section 4, we describe the difference-in-discontinuities
design we use to identify the causal impact of ESSER funds on district outcomes. We present our
results in Section 5 and conclude in Section 6.


2. Federal Funding of Public Schools


2.1 Funding Before the Pandemic
When compared to state and local sources, the federal government is typically a relatively minor
contributor to public K-12 school district budgets. The Title 1 program, authorized by the
Elementary and Secondary Education Act, is the largest federal program targeting local school

4
    The early theoretical foundations of this literature can be found in Bradford and Oates (1971a and 1971b).

                                                            6
districts, and it funnels most of its appropriations through the Part A funding formulas
(Congressional Research Service, 2017; Skinner, 2016). These four formulas disperse funds
across school districts and generally send more funds to districts that serve higher-need
populations (Congressional Research Service, 2017; Skinner, 2016; Gordon and Reber, 2020;
Gordon and Reber, 2023). Each formula allocates money to individual districts, and the sum of
the funds constitutes a district’s total amount of Title 1-A funding.5 During the pandemic,
Congress used the Title 1-A funding formulas to deliver emergency aid to school districts,
leading to an increase in the share of funding to schools from federal sources. 6
        The four formulas determine the Basic Grant, which allocates funds to all but about 70
school districts nationally, the Targeted Grant, the Education Finance Incentive Grant (EFIG),
and the Concentration Grant, which allocates funds to the largest and poorest districts. Each
formula takes a slightly modified version of the following form:


        𝑔
𝐹𝑢𝑛𝑑𝑠𝑖𝑡 = 𝑄 𝑔 (𝐹𝐶𝑖𝑡 , 𝑃𝑜𝑣𝑆ℎ𝑎𝑟𝑒𝑖𝑡 ) ×
                                                                    𝑔
max{𝑊 𝑔 (𝐹𝐶𝑖𝑡 , 𝑃𝑜𝑣𝑆ℎ𝑎𝑟𝑒𝑖𝑡 , 𝑆𝑡𝑎𝑡𝑒𝐹𝑎𝑐𝑡𝑜𝑟𝑠𝑠 ), 𝐻 𝑔 ( 𝐹𝑢𝑛𝑑𝑠𝑖𝑡−1 )}                                              (1)


where g denotes one of the four formulas, i indexes school districts, s indexes states, and t
indexes the fiscal year. Each formula has three main components: a qualification function 𝑄 𝑔 (⋅),
a weighting function 𝑊 𝑔 (⋅), and a hold harmless function 𝐻 𝑔 (⋅). In this paper we will focus on
exploiting the qualification dummy for two grant formulas, the Targeted and EFIG formulas,
which create small variations in Title 1-A funds that will map directly into large differences in
ESSER funds in the COVID-19 relief packages.
        The grant functions take four major inputs. The first two inputs are the formula child
count, 𝐹𝐶𝑖𝑡 , and the poverty share, 𝑃𝑜𝑣𝑆ℎ𝑎𝑟𝑒𝑖𝑡 , for a given district i in fiscal year t. The formula
child count is the number of children aged 5–17 assessed to be living in impoverished families,
in foster care, or who benefit from Temporary Assistance for Needy Families (TANF) (Skinner,
2016). The poverty share, 𝑃𝑜𝑣𝑆ℎ𝑎𝑟𝑒𝑖𝑡 , is the formula child count for a given school district
divided by the total number of children aged 5–17 living within that school district’s

5
  For a more complete overview of the funding formulas, see Skinner (2016), Congressional Research Service
(2017), and Gordon and Reber (2023).
6
  We show trends of school district revenues by source in Panel A of Appendix Figure 1 and per pupil revenues for
SY 2019 in Appendix Table 1.

                                                        7
boundaries.7 These inputs are estimated with the Small Area Income and Poverty Estimates
(SAIPE) for school districts from the United States Census Bureau (2023) using American
Community Survey data supplemented by tax data. The third input, 𝑆𝑡𝑎𝑡𝑒𝐹𝑎𝑐𝑡𝑜𝑟𝑠𝑠 , considers a
state’s own expenditures on education three years prior, equity in funding across school districts,
grant-specific state-level funding minima, and other grant-specific factors (Gordon and Reber,
2023). Finally, each formula has a hold-harmless provision, which takes into consideration the
                                                                     𝑔
amount of funds appropriated in the previous year, 𝐹𝑢𝑛𝑑𝑠𝑖𝑡−1, to ensure that no district receives
a large year-to-year decrease in funding. In most cases, hold-harmless provisions limit year-to
year-funding reductions to 15% of last year’s total, ensuring each district receives at least 85% of
the nominal funds it received the previous year (Skinner, 2016).
        For the Basic Grant, which made up 41.5% ($6.5B) of all Title 1-A funds in FY2017, the
qualification function requires a district to have at least 10 formula children and a 2% district
poverty share (Congressional Research Service, 2017). If a district qualifies for the basic grant,
the funding formula takes the number of formula children and scales them using state factors,
with each formula child receiving equal weight within a given state. A district that would see its
Basic Grant be reduced in a given year is entitled to no less than 85% of its previous fiscal year’s
basic grant allocation, so long as it still qualifies for the Basic Grant.
        The Concentration Grant, which distributed $1.4B (8.5% of all Title 1-A funds) in
FY2017, has two main differences from the Basic Grant (Congressional Research Service,
2017). First, the Concentration Grant requires districts to have either 6,500 formula children or a
15% poverty share to qualify for a grant. Second, a district that qualified for a Concentration
Grant in the previous year is entitled to hold-harmless funds even if it no longer qualifies for a
Concentration Grant (crossing the formula child or poverty share threshold) for up to four years.
        The other half of Title 1-A funds ($7.6B in FY2017) is appropriated through the EFIG
and Targeted Grant formulas (Congressional Research Service, 2017). To qualify for these grant
formulas, a district must have at least 10 formula children and a poverty share above 5%. In
these formulas, the weighting function for formula children is complex and rewards districts with
either high counts of formula children or high poverty shares (Gordon and Reber, 2023). For the
EFIG formula, the state factors include equity in funding across districts as well as state spending


7
 Note that these measures are estimates geographically and do not align with actual enrollment in the public school
district for which these measures are attached.

                                                         8
on education compared to state incomes (Skinner, 2016). Crossing the 5% poverty share and
qualifying for the EFIG and Targeted Grant formulas is the discontinuity that we use in our
difference-in-discontinuities research design, as discussed below.
         Each year, Title 1-A funds are determined through a multi-step process. First, Congress
sets funding levels for each of the four formulas for the year. Then, the Department of Education
calculates initial grants to school districts using the most recent SAIPE estimates for geographic
school districts. These allocations are then sent to state education agencies (SEAs), which are
allowed to reserve a small fraction of their state’s Title 1-A funds for approved purposes. The
state agencies allocate the remaining funds to their school districts (LEAs) following the same
Title 1-A grant formulas, making slight modifications to the allocations to accommodate charter
schools (Skinner, 2016).8


2.2 Funding During the Pandemic
In a typical recession, the budgets of state and local governments come under stress due to
borrowing constraints and balanced budget requirements. These constraints make federal aid to
sub-national governments an important policy tool during economic downturns (Clemens and
Veuger, 2021a; Congressional Research Service, 2009). In pre-pandemic recessions, federal aid
to K-12 public schools was included in spending packages such as the American Recovery and
Reinvestment Act. However, the pandemic-era Elementary and Secondary School Emergency
Relief Fund (ESSER) is unique in both the amount of funding Congress dedicated to school
districts and in Congress’ use of the existing Title 1-A formulas to determine funding levels
(Gordon and Reber 2020).9
         Between March 2020 and March 2021, Congress appropriated a total of about $190
billion to the ESSER program in the CARES Act, the Coronavirus Response and Relief
Supplemental Appropriations Act (CRRSAA), and the American Rescue Plan (ARP). These
funds were first given to state education agencies (SEAs) in proportion to their Title 1-A


8
  Explicitly calculating the state-level adjustments to accommodate public charter schools is not feasible with
publicly available national datasets. To do this final step of calculations would require not only knowing individual
charter school enrollments, but also the regular public-school districts in which those students live.
9
  As part of the American Recovery and Reinvestment Act, Congress delivered most of the $97 billion (nominal) aid
to education (including higher education) through a simple formula based on state population of persons aged 5 to
24 and total population. Many of the remaining appropriations were delivered through existing programs, but only
$10 billion (nominal) followed the Title 1-A formula (Congressional Research Service, 2009).

                                                          9
allocations. Then, the states were required to deliver no less than 90% of those funds to local
education agencies (LEAs, or school districts) in proportion to each district’s share of state Title
1-A funds from FY2020 (Skinner et al., 2022).10 Both the funding levels and inputs for each of
the FY2020 Title 1-A formula allocations were determined before the pandemic, largely by the
2018 SAIPE.11
        The ESSER program had many goals, largely dictated by the unique strains the pandemic
placed on school districts. These included the sudden adoption of remote learning and the
subsequent adaptation of district facilities to accommodate returns to in-person learning (Skinner
et al., 2022; Dewey et al., 2025; Alon et al., 2020; Alon et al., 2022; Halloran et al., 2021;
Halloran et al., 2023; Oster et al., 2021).12 ESSER funds were also supposed to help districts
make up for lost classroom hours due to closures, bolster mental health resources, and mitigate
layoffs, among other objectives (Skinner et al., 2022; Bryant et al., 2023).
        School districts faced few constraints in how they used ESSER funding, so long as funds
were deployed by legislatively specified deadlines. Initially, the $13B in ESSER I funding from
the CARES Act, the $54B in ESSER II funding from the CRRSAA, and the remaining $123B in
ARP ESSER funding needed to be spent by October 2021, October 2022, and October 2023
respectively. The 2023 deadline was extended by one year to 2024 (Skinner et al., 2022). A 2023
McKinsey survey of 487 school districts suggested that as much as $90B in ESSER funds was
left unspent as of June 2023 (Bryant et al., 2023).


3. Data
To assess the impacts of ESSER funds on school (district) performance during and after the
pandemic, we build an unbalanced panel dataset from the more than 17,400 public school
districts in the Common Core of Data (CCD) district fiscal13 and non-fiscal files for school years
2010 to 2023 (Institute of Education Sciences (IES), 2024b).14 We attach to these data the annual

10
   ESSER I used Title 1-A funds from FY2019 and ESSER II and ARP ESSER used FY2020 allocations. We use
the FY2020 formulas to create our research design since the vast majority of ESSER funds were delivered through
ESSER II and ARP ESSER.
11
   In Appendix Table 2, we present a timeline of events relevant for the ESSER program and school districts during
COVID-19.
12
   Closures also had an important role in shaping the pandemic’s impact on labor markets (Alon et al., 2020; Alon et
al., 2022).
13
   We deflate all fiscal variables by the Personal Consumption Expenditures Chain-Type Price Index, which we
access through FRED (BEA, 2024).
14
   We access these data using the Urban Institute’s (2024) Education Data Explorer where possible.

                                                         10
SAIPE estimates from the US Census Bureau (2023) because formula child count and poverty
share estimates are key inputs into the Title 1-A funding formulas. Roughly 12,700 school
districts have estimates from the SAIPE and data from the CCD. The vast majority of these
districts are categorized as traditional geographic school districts. Many charter schools do not
receive poverty estimates from the SAIPE as they may lack traditional geographic boundaries,
which means we cannot include them in our analysis even though they received federal funds
during COVID. We further restrict our panel to districts with non-zero enrollment, and with
complete data for total federal revenues, enrollment, and SAIPE poverty estimates between 2017
and 2022.15
         We supplement the fiscal, employment, enrollment, and district poverty data from the
CCD and SAIPE with data on ESSER allocations to school districts from the Department of
Education’s Education Stabilization Fund API (US Department of Education, 2024). We
combine several iterations of the API to build a dataset of ESSER allocations that includes
$170B in funds.16 These ESSER amounts are reported to the Department of Education by the
SEAs on behalf of their districts as part of federal grant reporting guidelines.
         In addition to enrollment, staffing, expenditure, revenue, and poverty estimates, we are
interested in how school districts performed in terms of student achievement during the
pandemic. We use harmonized test score data from the Stanford Education Data Archive
(SEDA), which uses the National Assessment of Educational Progress (NAEP) to make spring
state-level testing of 3rd to 8th graders comparable across states and years. These data scale state
test scores to the 2019 NAEP distribution, putting each district’s score in terms of 2019 standard
deviations and demeaning each score by the 2019 average (Reardon, et al., 2024). We are able to
match just over 7,000 districts with complete SEDA testing data for reading and math.17
         Next, we include additional variables to understand how districts performed during the
pandemic. First, we add data on chronic absenteeism from Ed Data Express (2024), which the
Biden administration identified as a key driver of low test performance during the pandemic

15
   In SY 2019-2020, all districts in California were attributed 0 teachers and staff employed in the CCD files. We
have replaced these missing observations with their values from the previous school year. Staffing counts include
teachers as a sub-category. For regressions that compare changes in staff and teacher counts, we drop districts that
report teacher counts in excess of their total staff counts, as they are likely misreported.
16
   We combine multiple data vintages to build a DUNS ID to NCES ID crosswalk that allows us to capture districts
that come in and out of the API vintages across pulls, using the most recent value available for each district’s final
allocation of ESSER Funds. We pulled the API on 12/1/2022, 10/19/2023, and 2/4/2024.
17
   Naturally, districts that do not enroll 3rd through 8th graders will not show up in the SEDA test score data.

                                                          11
(CEA, 2023). These data count students who are absent for at least 10% of school days. Second,
we analyze a cellphone-based mobility measure from Parolin and Lee (2022), which measures
district-level declines in cellphone visits around schools for calendar years (CY) 2020, 2021, and
2022 compared to their same month in 2019. We average these data by calendar year for each
school district.18 Finally, we include a measure of community average daily case rates around
school districts from the COVID School Data Hub (CSDH). The CSDH apportions average daily
county-level COVID cases per 100,000 residents to zip codes by population and then attaches
these measures to school districts (CSDH, 2023). We average each district’s daily case rate
across reporting weeks within SY 2021, the only year for which we have data. While all districts
in sample have CSDH case rate data, only 2,915 districts have complete mobility data.
        Our main explanatory variable is ESSER funds per pupil, where the total amount of
ESSER funds awarded across all three installments is summed and divided by a district’s fall
2019 enrollment total (SY 2020).19 Throughout this paper, outcomes measured on a per pupil
basis use the SY 2020 enrollment estimates, taken in the fall of 2019. For event study style
presentations, we will use total district federal revenues per pupil before the pandemic and then
add ESSER funds to create a composite federal funding per pupil measure in 2021.20
        Given that many variables in our dataset have imperfect coverage, we include districts as
data are available. This is why we describe our dataset as an unbalanced panel. Data availability
also explains why observation counts differ across presented regressions even when these
regressions analyze samples within the same bandwidth from the cutoff. We provide summary
statistics for each year of data in our sample in Appendix Table 3 as well as a list of variables,
along with descriptions, measurement frequencies, and sources in Appendix Table 4.


4. Methodology


4.1 The Difference-in-Discontinuities Design


18
   We also corroborate these data with shares of the school year by learning mode (in-person, hybrid, or virtual) as
reported during the 2020-2021 school year through the COVID School Data Hub (CSDH, 2023).
19
   We drop the top 0.01% of ESSER funds per pupil because the amounts are implausible on a per pupil basis.
Additionally, some districts reported negative ESSER amounts, which we corrected by removing the errant negative
sign.
20
   This also requires using the special CCD exhibits to remove ESSER funds from the post-COVID federal revenues
series so as not to double count funds.

                                                         12
ESSER funds are likely endogenous to school district outcomes. District poverty rates are closely
related both to educational outcomes and to the allocation of ESSER funds (Hanushek et al.,
2019). Figure 1 shows that while the pandemic reduced test scores in both reading and math
across the board, the size of the decline was closely linked to district poverty shares. Moreover,
the ESSER funding allocation process allowed states some opportunities to change the amount of
funds received by districts. States were able to change ESSER allocations by withholding up to
10% of funds for their own use and by approving and making allocations to new charter schools
(Skinner and Dortch, 2020; Office of Elementary and Secondary Education, 2020). School
districts could also choose not to accept ESSER funds for unobserved reasons. The endogeneity
of ESSER funds and omitted variables likely bias a simple OLS regression of outcomes on
ESSER funds.
       The potential sources of bias associated with a naïve regression of ESSER funds on
school district performance measures motivate our use of a discontinuity in the Title 1-A formula
as a source of identification within a difference-in-discontinuities research design. Districts that
just qualified for additional Title 1-A and ESSER funds and districts that did not were likely to
have continued on similar trends in the absence of the additional funds. We exploit this to
estimate a local average treatment effect for the impact of ESSER funds on school performance
and budget outcome variables for districts around the 5% poverty share threshold.
       The difference-in-discontinuities design exploits the qualification functions for the EFIG
and Targeted Grants. To qualify for funds from the EFIG and Targeted Grant formulas in
FY2020, a district had to have an area poverty rate of at least 5% among children aged 5–17 in
the 2018 SAIPE estimates. The ESSER funds were allocated in SY 2020 and SY 2021 to school
districts based on the districts’ shares of state Title 1-A funds from FY2020. As districts below
the 5% cutoff only received ESSER funds in proportion to their Basic and Concentration Grant
allocations, whereas districts above it received ESSER funds in proportion to their funds from all
four formulas, a discontinuity in ESSER funds should form around the cutoff.
       To implement the difference-in-discontinuities design, we first construct our running
variable, the distance from the 5% poverty cutoff measured in formula children, which we term
“centered formula children”:



                                                 13
𝑅𝑖 = 𝐹𝐶2018𝑖 − (𝐶ℎ𝑖𝑙𝑑𝑃𝑜𝑝2018𝑖 × 0.05).                                                                        (2)


𝐹𝐶2018𝑖 is the formula child count, i.e., children aged 5–17 estimated to be in poverty in the 2018
SAIPE, within school district i, and 𝐶ℎ𝑖𝑙𝑑𝑃𝑜𝑝2018𝑖 is the total number of children aged 5–17
within district i in the 2018 SAIPE. By multiplying the total number of children aged 5–17 by
0.05, we calculate the 5% poverty cutoff for each school district.
        In Figure 2 we show the relationship between our “centered formula children” variable
and both pre-pandemic federal funds and ESSER funds per pupil for districts within 100 formula
children of the cutoff. In Panel A we see a discontinuity of about $135 per pupil in pre-COVID
federal funds at the 5% poverty threshold, where the qualification cutoff is marked with a
vertical line. This motivates using the difference-in-discontinuity approach as opposed to a
simple regression discontinuity design.21 In Panels B and C, we see a similarly clear increase in
average ESSER funds per pupil right at the threshold, with all districts on display in Panel B and
districts binned in Panel C.22 Here the discontinuity amounts to around $400 per pupil. In Panel
D, we show an event study using a series of total federal funds, including ESSER. We can see
that the pre-pandemic difference in Title 1-A funds is stable over-time, and we see a strong
increase of around $400 per pupil from ESSER funds during the pandemic.
        As the 5% poverty threshold existed prior to the pandemic, we analyze the arrival of the
ESSER funds using a difference-in-discontinuities design (Grembi et al., 2016). We estimate
regressions of the following form, where i indexes school districts and t indexes time periods:


𝐸𝑆𝑆𝐸𝑅𝑖𝑡 = 𝛼 + ∑𝑡≠𝐵𝑎𝑠𝑒[𝛽𝑡 (𝐷𝑖 ∙ 𝐼{𝑌𝑒𝑎𝑟 = 𝑡}𝑡 ) + 𝛾𝑡 (𝑅𝑖 ∙ 𝐼{𝑌𝑒𝑎𝑟 = 𝑡}𝑡 ) + 𝜃𝑡 (𝐷𝑖 ∙ 𝑅𝑖 ∙
𝐼{𝑌𝑒𝑎𝑟 = 𝑡}𝑡 )] + 𝛿𝑖 + 𝜏𝑡 + 𝜀𝑖𝑡                                                                              (3a)
𝑌𝑖𝑡 = 𝛼 + ∑𝑡≠𝐵𝑎𝑠𝑒[𝛽𝑡 (𝐷𝑖 ∙ 𝐼{𝑌𝑒𝑎𝑟 = 𝑡}𝑡 ) + 𝛾𝑡 (𝑅𝑖 ∙ 𝐼{𝑌𝑒𝑎𝑟 = 𝑡}𝑡 ) + 𝜃𝑡 (𝐷𝑖 ∙ 𝑅𝑖 ∙
𝐼{𝑌𝑒𝑎𝑟 = 𝑡}𝑡 )] + 𝛿𝑖 + 𝜏𝑡 + 𝜀𝑖𝑡 ,                                                                            (3b)


Equation 3a is the first-stage regression equation used to identify the amount of ESSER funds
per pupil associated with the 5% poverty threshold. Our parameters of interest are the post-


21
  Panels A, C, and D of Figure 2 weight bins by SY 2020 (fall 2019) enrollments.
22
  Following the visual inference recommendations from Korting et al. (2023), we present our discontinuity figures
without trend lines.

                                                        14
COVID values of 𝛽𝑡 , which capture the impact of crossing the 5% poverty threshold on ESSER
funds. We capture the impact of crossing the 5% poverty threshold and qualifying for additional
funds with the dummy variable 𝐷𝑖 , which takes a value of 1 if a district has crossed the poverty
share threshold and 0 if it has not crossed the threshold. We will refer to 𝐷𝑖 as the qualification
dummy in our tables. We will also present a combined coefficient in our tables, which is the
linear combination of all post-COVID values of 𝛽𝑡 . The running variable, 𝑅𝑖 , is the centered
formula child count as described in Equation 2. We interact the qualification dummy and the
running variable with a set of year dummies, 𝐼{𝑌𝑒𝑎𝑟 = 𝑡}𝑡 , which take a value of 1 in each
included year and 0 otherwise, with the last available pre-COVID year of data left out as the base
year. Given the availability and measurement periods of our datasets, the base year varies
depending on the outcome variable.23 These equations use a first-degree polynomial of the
running variable, but we will present specifications using a second-degree polynomial as well.
We include district and time fixed effects, shown as 𝛿𝑖 and 𝜏𝑡 , respectively. Finally, 𝜀𝑖 is an error
term. We weight these regressions by district enrollment in fall 2019. Our robust standard errors
are clustered at the district level since qualification for additional ESSER funds is determined
district-by-district (Abadie et al., 2020; Abadie et al., 2023). We use bandwidths ranging from 45
centered formula children to 55 centered formula children from the poverty threshold. The
sample of districts within 50 formula children of the threshold, which is our preferred
bandwidth.24
         Equation 3b is the reduced-form equation. In this equation, we replace the left-hand side
variable from Equation 3a, ESSER funds per pupil, with our outcome of interest, 𝑌𝑖𝑡 . We use
equations 3a and 3b to estimate two objects of interest, the intent to treat (ITT) effect of
qualifying for additional ESSER funds and the local average treatment effect (LATE) associated
with additional ESSER funds. The post-COVID estimates of 𝛽𝑡 in equation 3b show the
relationship between qualifying for additional ESSER funds and changes in school district
outcomes. These are our estimates of the ITT effect. Additionally, we can recover a LATE by
scaling the post-COVID 𝛽𝑡 estimates in equation 3b by the first-stage estimates from equation

23
   For example, absenteeism is measured throughout the 2019-2020 school year, so any observation that overlaps
with COVID is counted as in the “post” period. This explains why some outcomes have different numbers of pre-
trend periods than other outcomes.
24
   Our sample does not capture the effects of ESSER funds in the South as many districts are either too large or too
poor (or both) to be close enough to the 5% poverty cutoff. We provide a map of our 50 formula child sample in
Appendix Figure 2.

                                                         15
3a. The LATE gives the impact of additional ESSER funds on district outcomes. However, the
LATE requires the assumption that qualifying for additional ESSER funds only impacts district
outcomes through receiving ESSER funds.
       Where appropriate, we also use a simple cross-sectional regression discontinuity
specification. We do this when investigating how ESSER funds jump across the 5% poverty
threshold, when estimating COVID outcomes for which there are no pre-COVID data, or for
graphing results as changes from their pre-COVID baselines. These regressions take the
following form:


𝑌𝑖 = 𝛼 + 𝛽𝐷𝑖 + 𝛾𝑅𝑖 + 𝜃(𝑅𝑖 ∙ 𝐷𝑖 ) + 𝜀𝑖 ;                                                          (4)


where i indexes school districts. 𝐷𝑖 is a dummy for qualifying for additional ESSER funds by
having a poverty share over the 5% threshold in the 2018 SAIPE, and 𝑅𝑖 is our running variable
of centered formula children from the 5% poverty share threshold. Here, we do not interact our
right-hand side variables with a series of year dummies; rather, we run these regressions cross-
sectionally year-by-year. Outcomes in this specification can be either observations for a specific
year or changes from a pre-COVID baseline. Just as in Equations 3a and 3b, we can also include
a second-degree polynomial in our running variable. We present robust standard errors for this
specification and weight observations by enrollments in SY 2020. 𝜀𝑖 is an error term.
       A limitation of our discontinuity-driven designs is that they produce estimates of
treatment effects local to the 5% poverty qualification cutoff. In Appendix Table 5, we show
how the typical school district in our preferred bandwidth sample, 50 formula children from the
cutoff, compares to the universe of school districts from the CCD data. We can see that the
typical student in a school included in our difference-in-discontinuities sample attends a district
that is smaller in terms of both students and number of schools, as well as significantly higher in
socio-economic status than districts outside of our sample. Naturally, students in our difference-
in-discontinuity sample attend schools in districts with an average poverty share of around 5%,
whereas the typical student nationally attends a district with a poverty share just under 20%.


4.2 Diagnostics and First-Stage Results



                                                 16
The difference-in-discontinuities design relies on several assumptions. First, the discontinuity
needs to be associated with a strong jump in ESSER funds per pupil, which we confirm in Figure
2 and Appendix Table 6. Panel A of Figure 2 shows a pre-existing jump in Title 1-A funds
across the qualification threshold of about $135 per pupil. This jump then expands to around
$400 per pupil, as shown in the un-binned scatterplot in Panel B and the binned scatterplot in
Panel C of Figure 2. In Panel D of Figure 2, we present the event-study version of our first-stage
results. This figure shows the coefficients on the qualification dummy for additional ESSER
funds, interacted with dummies for each year, with combined federal and ESSER funds per pupil
as the dependent variable. Here, we leave out SY 2020 as the base year. In the pre-COVID
periods, the $135 per pupil gap in federal funding appears to be stable. However, in the post-
COVID period, districts that qualified for additional ESSER funds saw a roughly $400 per pupil
increase in their combined federal funding, consistent with our estimates from Appendix Table
6.25 The jump in ESSER funds at the 5% poverty threshold represents a one-time 11% increase
in discretionary instructional funding for the typical district in this sample.
         The second assumption is that districts close to and on both sides of the 5% poverty
threshold would have remained on parallel trends in the absence of the ESSER program. If this
assumption holds, we would expect the covariates of districts just on either side of the
qualification threshold to be on parallel trends before COVID, even if the pre-existing
discontinuity in federal funding generates differences in levels.
         To test this assumption, we look at covariate balance in trends. For this analysis, we place
covariate trends on the left-hand side of Equation 4. Specifically, we show estimates using the
change in absenteeism from SY 2017 to SY 2019, percentage change in enrollment from fall
2016 to 2019, percent change in total revenues per pupil from SY 2017 to SY 2019, percent
change in federal revenues per pupil from SY 2017 to SY 2019, change in the share of
enrollment that is White from fall 2016 to 2019, change in the share of enrollment that is Black
from fall 2016 to 2019, changes in reading scores from spring 2017 to 2019 measured in 2019
standard deviations, changes in math scores from spring 2017 to 2019 measured in 2019 standard


25
  We show a range of specifications for our first stage in Appendix Table 6. Under our preferred specification for
the first stage, which uses a 50-formula child bandwidth and a first-degree polynomial, crossing the 5% poverty
threshold results in an estimated additional $388 per pupil in ESSER funding, as shown in Column 3. This estimate
has an F-statistic of 34. Across the six specifications shown in Appendix Table 6, none has an F-statistic below 17 or
an estimate of the additional ESSER funds of under $350 per pupil.

                                                         17
deviations, changes in teachers per pupil from fall 2016 to 2019, and changes in staff per pupil
from fall 2016 to 2019, where per pupil outcomes use fall 2019 enrollments.
         The results are reported in Panel A of Appendix Figure 3 and in Appendix Table 7.
Broadly, these results support our difference-in-discontinuities approach. Only a single trend
showed a statistically significant difference across the threshold: the share of white students.
However, the scatter plots reveal that this trend is driven by outlier districts away from the
qualification threshold.26
         Next, we describe how levels of covariates change across the cutoff in Panel B of
Appendix Figure 3 and in Appendix Table 8.27 We find evidence of small differences that are
likely directly associated with the Title 1-A program, which aims to increase teacher counts.
Indeed, this is why we adopt a difference-in-discontinuities design, which differences out these
baseline gaps in levels, rather than a static regression discontinuity design.


5. Results
This section presents estimates of the effects of additional ESSER funds on school district
performance and decisions during the pandemic. To establish that the qualification discontinuity
causes an outcome difference, we rely on three complementary pieces of evidence. First, the
regression tables must show a statistically significant coefficient on the qualification dummy.
Second, event study figures need to exhibit flat pre-trends whenever pre-COVID data are


26
   Scatterplots of trend changes around the cutoff are displayed in Appendix Figure 3 Panel A. Each cell of
Appendix Table 7 shows the coefficient on the qualification dummy for crossing the 5% poverty threshold and
qualifying for extra funds for a different bandwidth and outcome variable. We present results using the 45, 50, and
55 student bandwidths around the qualification cutoff in separate rows. In our preferred specification, a 50-formula
child bandwidth and a first-degree polynomial, we see that only the share of White students seems to exhibit a slight
differential trend when comparing changes in districts just above the qualification cutoff and just below. We show
estimates using a second-degree polynomial in Panel B of Appendix Table 7. These estimates show qualitatively
similar results to those using our preferred polynomial choice in Panel A of Appendix Table 7.
27
   In Panel B of Appendix Figure 3 and in Appendix Table 8, we show estimates of Equation 4 that replace ESSER
funds per pupil with potential covariates in levels on the left-hand side. Across districts immediately surrounding the
cutoff, both the enrollment level and the White student share are smooth in levels. We observe statistically notable,
but economically modest, discontinuities with respect to baseline levels of federal revenues and teachers per pupil.
These discontinuities in baseline levels are in line with what one should expect as a consequence of the Title 1-A
formulas. The discontinuity in the baseline level of teachers per pupil is 0.005, implying 1 additional teacher per
every 200 students. In 2019, the typical district in our sample had about 14 teachers for every 200 students. We note
that enrollment and the share of students who are White exhibit economically non-trivial estimated differences
around the cutoff, though the estimates are noisy. Visual inspection of Panel B of Appendix Figure 3 shows that
these estimates are driven in large part by a district located three bins to the left of the cutoff, which is an outlier
with respect to both enrollment and the share of students who are White. Our full set of results is robust to dropping
this outlying district, Leander Independent School District in Texas.

                                                          18
available. Third, binned-scatter plots centered on the qualification threshold should reveal a
discrete jump in the outcome of interest. Because we examine a wide range of outcomes, we do
not present all three diagnostics in the body of the text, though we generally provide graphical
evidence in the appendix.


5.1 District Performance During the Pandemic
As we saw in Figure 1, test scores declined during the pandemic. Multiple sets of researchers
have associated these decreased test scores with virtual learning during SY 2020 and SY
2021(Halloran et al., 2021; Oster et al. 2021; Goldhaber et al., 2023). Enrollment also declined
sharply. Nationally, about 1.25 million students were “missing” from public schools in SY 2021
compared to SY 2020 (measured in the fall semester). By SY 2023, enrollment had only
recovered to a 750,000 student decrease compared to SY 2020. Students in grades K-8 saw
larger trend declines in enrollment than high school students.28 Against this backdrop we
investigate if the funds delivered through ESSER allowed schools to alleviate the decrease in test
scores and to retain additional students.


5.1.1 Traditional Measures of School Performance


         In Table 1, we report estimates for our main district performance outcomes. For each
outcome variable, we show estimates for our preferred specification. These regressions use a first
degree polynomial and a 50 formula child bandwidth around the qualification threshold.29
Columns 1 through 6 show how test scores for math and reading evolved in response to the
additional $388 per student in ESSER funds. Regardless of bandwidth choice, we see no
evidence that ESSER funds had an effect on test scores during SY 2022 and SY 2023.30 Using
our math scores coefficient for 2023 in Column 2, we can rule out effects larger than a 0.018
(2019) standard deviation increase in scores from ESSER funds. Math test scores declined about


28
   In Panel A of Appendix Figure 4, we show national trends in enrollment in levels. In Panel B of Appendix Figure
4, we present a more detailed look at enrollment changes across grade-level groupings, indexing enrollment changes
for grades K-5, grades 6-8, and grades 9-12 to their pre-COVID values.
29
   We present accompanying tables of a full set of first- and second-degree polynomial estimates at alternative
bandwidths in Appendix Tables 9 and 10.
30
   No testing was conducted in 2021. The baseline for these regressions is the last available testing year in the Spring
of 2019 (SY 2019).

                                                          19
0.1 standard deviations from SY 2019, which means we can rule out that the $388 per student in
ESSER funds addressed more than 18% of the decline in scores. For reading scores, the
precision of our estimates only allows us to rule out effects larger than 0.038 standard deviations
as a result of incremental ESSER funds using the estimates in Column 5. The maximum effect
size for reading that we can rule out is a much larger share of the 0.05 standard deviations of
learning loss from SY 2023 compared to SY 2019. We present event study versions of these
estimates in Panel A and B of Appendix Figure 5.31 Overall, our point estimates center around
zero impacts of ESSER funds on test scores, but our standard errors do not allow us to rule out
meaningful effect sizes that would be consistent with the broader literature on school quality
inputs (Jackson and Mackevicius, 2021).
         We present our results for enrollment in Columns 7 through 9 of Table 1.32 Our preferred
specification in Column 8 shows that an additional $388 per pupil in ESSER funds helped retain
about 125 students on average in SY 2021. The point estimate remains positive but attenuates in
magnitude in SY 2022 and SY 2023, consistent with enrollment rebounds exhibiting a moderate
degree of “catch up” in districts to the left of the discontinuity over time. Aggregating the
coefficients for SY 2021, SY 2022, and SY 2023 shows a statistically significant total retention
of about 283 student-years over the three school years. These estimates suggest that 2.3% of pre-
COVID enrollment was retained by the $388 per student in ESSER funds, or about 1.9 student-
years per $10,000 of ESSER funds. Panel C of Appendix Figure 5 shows these estimates as an
event study.33 These results suggest that ESSER may have helped schools retain students, and
that in the absence of ESSER funds enrollment declines might have been even higher than those
observed. Enrollment declines were about 5% for the median district in 2022 compared to 2019
and about 2% nationally (Burtis and Goulas, 2023).
         We show the impact of additional ESSER funds on enrollment by grade in Table 2. These
regressions use our preferred 50 formula child bandwidth and first-degree polynomial
specification. First, in Column 1, we show the same estimate as that in Column 8 of Table 1 for


31
   A large driver of the imprecision in these estimates is the decline in sample available, as many districts do not
have SEDA estimates, or are high schools that do not partake in the underlying testing.
32
   Since enrollment is measured in the fall semester, we use SY 2020 as the base year for these regressions.
33
   Appendix Figure 6 shows changes in enrollments from SY 2020 (fall 2019) for SY 2021, 2022, and 2023 against
our centered running variable of formula children, weighting our bins by pre-COVID enrollments. We see a clear
discontinuity in enrollment changes around the cutoff in 2021 and 2022, and an attenuated change in 2023,
consistent with the estimates reported in Table 1.

                                                         20
total enrollment changes. Then, we present estimates for districts that had any enrollment at a
given grade-level group in 2019 in Columns 2 through 6. We present estimates for pre-
kindergarten, kindergarten, grades 1–5, grades 6–8, and finally grades 9–12, with the latter three
categories typically corresponding to elementary, middle, and high school students. We can see
from Columns 3, 4, and 5 that most of the 283 student-year increase came from movement in
lower-grade levels, especially grades 1 through 8. We find no evidence that qualifying for
additional ESSER funds increased enrollments for high school students.
         Even if students are enrolled, they may attend class at lower rates as a result of the
impacts of the pandemic. We display our measure of the impact of ESSER funds on absenteeism,
which is defined as a student missing 10% of school days or more, in the final three columns of
Table 1. Across specifications, we find that qualifying for additional ESSER funds had a limited
effect on absenteeism in SY 2020 and 2021, but may have reduced absenteeism in SY 2023,
although these estimates are individually imprecise. However, the combined coefficient shows
that ESSER funds likely decreased the number of absentee students by about 220 absentee
students-years over the three years. The average school in our sample had about 337 absentee
students annually before COVID-19. These results suggest that declines in test scores associated
with ESSER may be the result of retaining lower-performing students, as schools with less
absenteeism may have seen their test scores suffer from negative selection into their test-taking
student population.34
         In order to probe the possibility of negative selection in the population of students taking
standardized tests, we investigate how the population of test takers changed in response to
ESSER funds in Appendix Table 11.35 Overall, it seems unlikely that negative selection can
explain the negative point estimates for the impact of ESSER funds on test scores.36


34
   This is evidence that suggests increased absenteeism would help, rather than hurt test scores. These results are in
conflict with the conclusions of the Biden CEA, which claimed that absenteeism was a driver of low test scores
(CEA, 2023).
35
   We use test score data from Halloran et al. (2023), which includes counts of the test taking population for each
district. We first show how the total number of test takers changed in 2021, 2022, and 2023 in Column 1. We find
suggestive evidence that the count of test takers increased in 2021, but these estimates are imprecise. In Columns 2
and 3, we show how counts of socioeconomically advantaged and not socioeconomically advantaged test takers
changed in response to qualifying for additional ESSER funds. In Column 3, we can see a statistically significant
increase of about 230 socioeconomically disadvantaged test takers in 2022. However, Column 4 shows that this
increase is a very modest share of the overall testing population.
36
   A second channel of interest may be that non-disadvantaged students left public schools for private schools. Using
data on private school enrollments in 2020 and 2022 from the private school universe survey (Institute of Education
Sciences, 2024b), we find suggestive evidence that private school enrollment within 5 miles of a school district that

                                                         21
5.1.2 Pandemic-Specific Measures of Performance
We next consider two outcomes of unique importance to the pandemic: the speed at which
school districts reopened and their potential contribution to the spread of COVID-19. School
districts that offered more days of in-person learning had higher mobility levels as measured by
cellphone activity, relative to their 2019 baseline (Parolin and Lee, 2022). Given that these
cellphone mobility data come already differenced and without pre-trends, we present them in
cross sections by year following Equation 4.
         Table 3 displays our estimates of the impact of ESSER funds on mobility. We show
results separately for each year of data and for both first- and second-degree polynomials. We
find that additional ESSER funds may have helped districts bring students back to the classroom
at a higher rate. In 2020, districts receiving additional ESSER funds had about a 5% increase in
mobility around their schools compared to mobility changes in below-cutoff districts. In 2021,
we see this gap increase to almost 9%. Finally, the mobility gaps persist into 2022, as districts
that received more funding appear to have had almost 11% more cellphone mobility compared to
their 2019 baseline measures than districts below the 5% poverty threshold.37
         Many observers and stakeholders were concerned that opening schools would increase
COVID-19 cases in the neighboring populations. In Appendix Table 13, we investigate whether
or not additional ESSER funding caused excess community spread of COVID-19, as measured
by average daily case rates in SY 2021 in the surrounding area.38 In our preferred specification,
we can rule out effects larger than an increase of half a case per 100,000 residents per day. This
upper bound can be compared to a control mean for districts below the qualification threshold of
about 24 cases per 100,000 residents per day.39


qualified for additional ESSER funds decreased slightly compared to 2020, and that private school enrollments 5 to
10 miles from a district that qualified for additional ESSER funds increased slightly. These results are shown in
Appendix Table 12. Although these results are statistically imprecise, the confidence intervals and point estimates
include estimates that would be complementary to our overall enrollment finding in terms of magnitudes.
37
   We show scatterplots with mobility changes against the running variable in enrollment weighted bins in Appendix
Figure 7. Here we can clearly see the discontinuity emerge and grow in each subsequent year, which corroborates
our estimates in Table 3.
38
   We present estimates of Equation 4 for three bandwidths and both polynomial specifications. We find that ESSER
funds had no detectable impact on case rates, and the zeros appear to be relatively precise.
39
   Using alternative case rates from the New York Times (2023), we find broadly similar results to those shown in
Appendix Table 13. It is important to note that these results only suggest that daily case rates and increased mobility
around schools do not appear to be correlated. Our research design cannot determine that school openings caused or
did not cause an uptick in COVID cases. Our difference-in-discontinuities design can identify the causal effect of

                                                          22
5.2 Measuring School District Choices
We now turn to estimates of how additional ESSER funds impacted school district choices
pertaining to staffing and district budgets. These results may suggest mechanisms through which
districts were able to retain students and return to in-person learning. For both staffing and
budget outcomes, we return to presenting full event study estimates from Equation 3b since
several years of pre-COVID data are available in our dataset.


5.2.1 District Staffing
We show the estimated impact of an additional $388 per pupil of ESSER funds on district full-
time-equivalent staffing levels per student in Table 4. Throughout the table we use our preferred
specification: a first-degree polynomial and a bandwidth of 50 formula children from the
qualification cutoff at the 5% poverty share threshold. In Column 1, we show estimates for
changes in total staff per pupil. We then break total staff into two mutually exclusive groups,
teachers and non-teachers, and show estimates for teachers per pupil in Column 2 and non-
teachers per pupil in Column 3.40 In the remaining three columns we include categories of staff
that may be of particular interest during the pandemic and with respect to federal grants:
counselors, instructional aides, and district administrators. Our estimates of the impact of
qualifying for additional ESSER funds on total staff, teachers, and non-teachers do not show
meaningful changes in Column 1 through 3. For total staff changes per pupil, we can rule out
combined increases over the three school years greater than 16 staff members per 1,000 students
each year, which is around 11% of the pre-COVID baseline of 147 staff per 1,000 students.
Similarly, we do not see a large change in the numbers of counselors per pupil or instructional
aides in Columns 4 and 5. Increasing support staff was identified by California as a “top priority”
for use of ARP ESSER allocations in 2021, and these findings show that we cannot detect
increases in these staffing categories (California State Board of Education, 2021). Column 6
reports, however, that ESSER funds may have increased the number of district administrators.




additional ESSER funds on outcomes at our qualification cutoff, but we cannot untangle the causal relationship
between the outcome variables themselves.
40
   In order for Columns 2 and 3 to add to Column 1, we only include districts in which the number of teachers
reported is smaller than the number of total staff reported, which should include teachers as a sub-category.

                                                        23
5.2.2 District Finances
Our final set of outcomes are district level expenditures and revenues. In Panel A of Table 5 we
investigate the causal relationship between additional ESSER funds and school district revenue
sources.41 In Column 1, we show changes in total district revenues per pupil in response to an
additional $388 per pupil in ESSER funds. These results show a $544 per pupil decrease in
revenues across the three years that is not significant at conventional confidence levels. In order
to understand which revenues declined, we show separate estimates for federal revenues, state
revenues, and local revenues per pupil in Columns 2 through 4, respectively.42 Column 2
suggests that in SY 2021, districts that qualified for an additional $388 per pupil in ESSER funds
saw increases in their federal revenues compared to schools below the cutoff. The combined
coefficient suggests the above-threshold districts received an additional $261 per pupil over three
years, which is likely a partial accounting of ESSER funds. In Column 3 we show that districts
that qualified for additional ESSER dollars did not see a meaningful amount of additional
funding from the state, with the combined coefficient suggesting an effect size of $102 per pupil
over 3 years, which is less than 2% of the pre-COVID mean on an annual basis.
         The last major category of revenues is from local sources, which typically make up the
majority of school revenues. From Column 4, we see that an additional $388 per pupil in ESSER
funds is associated with a significant decrease in local revenues in SY 2021 and SY 2022. The
combined estimates suggest that over the three years, crossing the qualification cutoff for
additional ESSER funds is associated with a $907 per pupil decrease in local revenues, or about
8% of the pre-COVID average. We also show the decline in local revenues as an event study in
Figure 3, which includes a flat pre-COVID trend in local revenues per pupil across the
qualification threshold. We can further contextualize this local revenue decline by looking at
changes in property tax collections. In Column 5, we can see that about 80% of the revenue
decline came from decreases in revenues from property taxes, with property taxes decreasing by
$718 per pupil across the three years.




41
   Recall that all per pupil variables use enrollments from 2019 (SY 2020) to scale outcomes, so changes are driven
only by the numerator.
42
   In the Common Core of Data, federal revenues include COVID funds. However districts likely have not reported
all of their funds in these data. There is also a special COVID exhibit for ESSER funds, but these data do not match
the DOE transparency API, which is our source for ESSER funds.

                                                         24
        Home prices rose rapidly during the pandemic (S&P Dow Jones Indices LLC, 2025). As
tax assessments increase with home value appreciation, one would expect voters to ask for relief.
We show several pieces of evidence to suggest that these revenue declines could be the result of
rate cuts, or relative decreases compared to districts that did not qualify for additional funds and
chose to raise additional local revenues. First, in Figure 4, using data from the National Council
of State Legislatures (2024) we show that property taxes were an active margin of adjustment
during the pandemic. We present the fraction of states that implemented any increase or any
decrease (not mutually exclusive) between 2016 and 2019 and between 2020 and 2023. For both
periods, over 40% of states made some change to decrease property taxes. Additionally, we
located the budgets and board meeting minutes for several districts in our sample that qualified
for additional ESSER funds and found that they did in fact lower property tax rates during the
pandemic. These districts are Barbers Hill Independent School District in Texas, Peaster
Independent School District in Texas, and Lake Geneva School District in Wisconsin.43 In fact,
Wisconsin schools broadly seem to have used their ESSER funds explicitly as a temporary
budget stopgap to curb property tax increases (Wisconsin Policy Forum, 2021). To take one clear
example, Lake Geneva School District called for a referendum to increase taxes after their
ESSER funds ran out in 2024, touting to voters that they used these funds for routine budget
needs (Lake Geneva Schools, 2025). These anecdotes provide suggestive evidence that districts
may have lowered their property tax rates. Federal relief dollars effectively crowded out local
revenue in school budgets.
        We also provide two pieces of evidence that our results are likely not driven by changes
in state funding. In Column 3 of Table 5 Panel A, we can see that state revenues do not jump
with the qualification dummy during the pandemic, as our results are not significant at traditional
levels and relatively small in magnitude. In Column 6, we also show that state formula funds do
not show changes that jump at the 5% poverty threshold either after COVID. These formulas are
typically used to equalize funding across districts within a state.
        In Panel B of Table 5, we present estimates for the impact of qualifying for additional
ESSER funds on district expenditures per pupil. We focus on total expenditures, elementary and


43
   Barbers Hill ISD lowered its tax rates from 1.26% to 1.15% in 2020 (Barbers Hill ISD, 2025). Peaster ISD has
steadily lowered its total tax rate since FY2021, which is set the August before the Fiscal Year, from 1.372% down
to 1.1664% in 2024 (Peaster ISD, 2024). Lake Geneva decreased its tax rate from 4.29 mill in FY2019 to 3.76 mill
in FY2021 (Wisconsin Department of Public Instruction, 2019 and 2021).

                                                        25
secondary education expenditures, capital outlays, instructional spending (including salaries),
salaries of all employees, and benefits of all employees on a per pupil basis in Columns 1
through 6, respectively. Across expenditure categories we find no evidence of changes in
spending that are detectable at a conventional significance level.44 The flypaper effect would
suggest that revenues from ESSER should be associated with corresponding expenditure
increases, which we do not detect at traditional significance levels across expenditure categories
in Table 5 (Gramlich, 1977; Hines and Thaler, 1995; Fisher, 1982). Instead, the districts appear
to have reduced tax revenues, passing the additional funds on to voters as in Lutz (2010).
         Our estimates suggest that a gap between declining revenues and constant expenditures
may have emerged as a result of additional ESSER funds. This would suggest that districts
would have needed to raise debt or draw on assets to accommodate the imbalance. We provide
suggestive evidence in support of the idea that districts may have issued new debt or relied on
assets to accommodate local revenue reductions in Appendix Table 14, but the data quality is too
low to draw strong conclusions. 45 Even so, there is some additional support for the possibility
that school districts raised debt during the pandemic. Lieberman (2023) reports that school
districts, through bond elections, in 2021 and 2022 proposed and passed larger numbers of bonds
than in 2020. The 2,134 bonds proposed by districts in 2022 almost matches the previous high in
2018 of 2,163 bonds proposed.
         Finally, in Appendix Table 15 we explore whether or not compensation per employee
meaningfully changed with additional ESSER funds.46 We find that total compensation per staff


44
   Interestingly, we also see almost no changes in the share of school expenditures among major revenue categories
when we look at the full sample of districts, including instruction, capital outlays, and student support services, in
Panel B of Appendix Figure 1.
45
   In Appendix Table 14, we explore how districts may have funded their revenue reductions after COVID. We
present estimates of Equation 3b using the debt and asset variables available to us in the Common Core Data files.
The debt data in the CCD files do not include critical details about the maturity and interest rates associated with
school bonds. In Column 3, we show how the face value (par value) of total debt changed for districts that qualified
for additional ESSER dollars. Across the three years, we find that districts that qualified for additional ESSER funds
raised an additional $1,809 per student across the three years in debt. However, these estimates are noisy, likely in
part due to data quality issues. To give additional context for these estimates, we examine changes in the interest
payments made by districts in Column 4. The interest payment change should equal the change in the face value of
debt, multiplied by the coupon rate. Using the estimates in Columns 3 and 4, the interest payment and total debt
changes imply a coupon rate of about 11% on average for new debts during the pandemic. These rates would be in
the right tail of the distribution of rates for school district bonds issued between 1997 and 2017 found by Biasi et al.
(Forthcoming). Although a rate of 11% is not impossible, it likely reflects the lack of reliability of the CCD data.
46
   Appendix Table 15 reports estimates of Equation 3b, but the left-hand side variable is total compensation per
employee. We use benefits and salaries as our measure of compensation, and we present these estimates for our
three main staff categories: total staff, teachers, and non-teachers.

                                                          26
member (teachers and non-teachers) increased in 2020 by $2,098 per employee, or a 2.5%
increase over pre-COVID baseline compensation. Compensation may have remained elevated in
2021, but this result is imprecise. Then, in 2022, compensation returned to baseline.47 These
results suggest a change in the labor share of district revenues during the pandemic, although the
combined effects are imprecisely estimated.


5.3 Potential Mechanisms for District Policy Responses


Our core results suggest that qualifying for additional ESSER funds allowed school districts to
retain students, return to hybrid or in-person learning faster, and reduced local revenues relative
to similarly situated districts that did not receive additional funds. At first glance, these results
may be puzzling as enrollment retention and returning to in-person learning could be at odds
with a revenue cut. Moreover, interest groups such as administrators and teachers might capture
some of the funds as memoranda of understanding between unions and districts were likely
updated in response to the pandemic (Peeples, 2020).
        We provide evidence to support the idea that the policy mix that we find as a result of
ESSER funds likely reflects increased political engagement, which may have been enabled by
additional ESSER funds. We map the 2016 through 2024 responses of the Cooperative Election
Study (CES) to school districts in our difference-in-discontinuities 50 formula child bandwidth
sample (Schaffner et al., 2025).48 For each question, we calculate the response rates across
districts above and below the qualification cutoff, weighted by district enrollments. We use
questions that capture whether or not a respondent voted or intends to vote in the upcoming
election, attended a local school board or city council meetings, or contacted a public official in
the past year. We then calculate the parental share among respondents that participated in each of
these engagement activities. These data allow us to see if political engagement changed

47
   A similar pattern is reflected in Columns 2 and 3 for teachers and non-teachers. However, the increase in
compensation during the pandemic seems to be more pronounced for non-teachers, with a 10.6% increase in 2021
compared to the pre-COVID baseline that is statistically different than zero. Overall, the combined $5,894 in
additional compensation per employee across the 3 years roughly translates into a $289 per student per year increase
in expenditures using the pre-COVID student to staff ratio.
48
   The CES is a large representative (around 60,000 respondents) survey cast around early October of each election
year. To our knowledge, the CES is the only long-standing survey that has sufficient sample and asks engagement
questions about voting, local government meeting attendance, and contacting local officials with demographics
available to identify parents with children under age 18. For details, see the documentation in Schaffner et al.
(2025).

                                                         27
differentially during the pandemic in districts that qualified for additional ESSER funds
compared to districts that did not qualify for additional funds.
           Figure 5 presents percent changes in electoral engagement from 2018 for districts in our
difference-in-discontinuities sample.49 In the top row, we show changes for all respondents in
districts that qualified for additional ESSER funds in blue and districts that did not qualify for
additional funds in red. Overall voting doesn’t seem to have changed differentially across these
sets of districts. However, changes in attendance at local government meetings appears to be
higher in districts that received additional funds in 2020, 2022, and 2024, and a more modest
difference in changes in rates of contacting local officials emerges for districts above the
qualification threshold later in the sample period. The second row of Figure 5 shows the changes
from 2018 in the parental share of respondents that participated in each political activity. We find
the strongest evidence that changes in the parental share of voters were larger in 2020 for
districts that received additional ESSER funds, and these changes may have persisted in 2022
and 2024. We find similar patterns for changes in the parental share of meeting attendees and
public official contacts, but these results show significant overlap between the bootstrapped
confidence intervals of both sets of districts. Overall, these data provide suggestive evidence that
federal grant spending and related policy priorities reflected the preferences of local residents,
which is consistent with several theoretical and empirical results in the literature on
intergovernmental grants (Bradford and Oates, 1971a; Bradford and Oates, 1971b; Gordon,
2004; Lutz, 2010).


6. Conclusion
Emergency funding for K-12 public schools was an important component of recession spending
packages during both the Great Recession and COVID, involving $50 billion and $190 billion in
additional expenditures, respectively (Shores and Steinberg, 2022). In this paper we have
provided causal evidence that the assistance provided by ESSER to local school districts may
have helped them to retain students and re-open their schools faster. However, we see no
evidence that ESSER funds helped to mitigate learning loss, at least in the short run. A meta-
analysis by Jackson and Mackevicius (2021) suggests that an additional $1,000 per year of K-12



49
     These changes are weighted by enrollment. We provide 95% bootstrapped confidence intervals for both series.

                                                          28
school spending raises scores by 0.012 standard deviations, and we cannot rule out estimates of
that magnitude.
       In addition to our findings for school performance, we provide evidence on school district
finances. While we find no statistically significant evidence that districts increased expenditures
in SY 2021 or SY 2022, we do find evidence that districts that qualified for additional ESSER
funds had statistically significant reductions in local revenues that were in excess of their ESSER
funds. This pass through of federal funds may partially explain why ESSER had minimal
impacts on learning loss for districts around the 5% poverty threshold. Moreover, increased
political engagement, especially from parents, may have driven school boards to direct these
funds towards property tax reductions and increased re-opening rates.
       Our findings suggest that ESSER funds were successful in achieving some but not all of
their policy goals, with positive effects on school reopenings and the retention of student
enrollments. However, our findings apply specifically to districts in the neighborhood of the 5%
poverty threshold for qualifying for additional ESSER funds. These districts have lower poverty
rates than a typical U.S. school district and are smaller than the national average. Our results
suggest that smaller and richer school districts used additional ESSER funds largely to reduce
property tax burdens and open schools, in contrast to the predictions of the flypaper effect.




                                                 29
References

Abadie, Alberto, Susan Athey, Guido W Imbens, and Jeffrey M Wooldridge. 2020. “Sampling-
   Based versus Design-Based Uncertainty in Regression Analysis.” Econometrica 88(1): 265-
   296.
Abadie, Alberto, Susan Athey, Guido W Imbens, and Jeffrey M Wooldridge. 2023. “When
   Should You Adjust Standard Errors for Clustering?” The Quarterly Journal of Economics
   138(1): 1-35.
Alon, Titan, Matthias Doepke, Jane Olmstead-Rumsey, and Michele Tertilt. 2020. “The Impact
   of COVID-19 on Gender Equality.” NBER Working Paper No. 26947.
Alon, Titan, Sena Coskun, Matthias Doepke, David Koll, and Michele Tertilt. 2022. “From
   Mancession to Shecession: Women’s Employment in Regular and Pandemic Recessions.”
   NBER Macroeconomics Annual Volume 36: 83-151.
Autor, David, David Cho, Leland D. Crane, Mita Goldar, Byron Lutz, Joshua K. Montes,
   William B. Peterman, David D. Ratner, Daniel Villar Vallenas, and Ahu Yildirmaz. 2022a.
   “An Evaluation of the Paycheck Protection Program Using Administrative Payroll
   Microdata,” Journal of Public Economics 211.
Autor, David, David Cho, Leland D. Crane, Mita Goldar, Byron Lutz, Joshua K. Montes,
   William B. Peterman, David D. Ratner, Daniel Villar Vallenas, and Ahu Yildirmaz. 2022b.
   “The $800 Billion Paycheck Protection Program: Where Did the Money Go and Why Did it
   Go There?” Journal of Economic Perspectives 36(2): 55-80.
Barbers Hill ISD. 2025. Required Postings. Available at:
   https://www.bhisd.net/departments/financetax-office/required-postings.
Betts, Julian R., 1995. “Does school quality matter? Evidence from the National Longitudinal
   Survey of Youth.” The review of economics and statistics 77(2): 231-250.
Biasi, Barbara, Julien Lafortune, and David Schonholzer. Forthcoming. “What Works and For
   Whom? Effectiveness and Efficiency of School Capital Investments Across the U.S.”
   Quarterly Journal of Economics.
Bradford, David F. and Wallace E. Oates. 1971a. “The Analysis of Revenue Sharing in a New
   Approach to Collective Fiscal Decisions.” The Quarterly Journal of Economics 85(3): 416—
   439.
Bradford, David F. and Wallace E. Oates. 1971b. “Towards a Predictive Theory of
   Intergovernmental Grants.” AEA Papers and Proceedings 61(2): 440—448.
Bryant, Jake, Wayne Redmond, Neil Shelat, and Casey Willis. 2023. “Stimulus Funding
   Deadlines Loom: How are K-12 Schools Adjusting Their Priorities?” Mckinsey & Company.
   Available at: https://www.mckinsey.com/industries/education/our-insights/stimulus-funding-
   deadlines-loom-how-are-k-12-schools-adjusting-their-priorities.
Burtis, Eloise and Sofoklis Goulas. 2023. “Declining School Enrollment Since the Pandemic.”
   The Hamilton Project, The Brookings Institution. Available at:
   https://www.brookings.edu/wp-
   content/uploads/2023/10/20231012_THP_Enrollment_FullPaper.pdf.
California State Board of Education. 2021. “California State Plan for the American Rescue Plan
   Elementary and Secondary School Emergency Relief Fund.” United States Department of

                                              30
   Education, OMB Number 1810-0754. Available at:
   https://oese.ed.gov/files/2021/08/California-ARP-ESSER-State-Plan.pdf.
Carbonari, Maria V., Miles Davison, Michael DeArmond, Daniel Dewey, Elise Dizon-Ross, Dan
   Goldhaber, Ayesha K. Hashim, Thomas J. Kane, Andrew McEachin, Emily Morton, Atsuko
   Muroga, Tyler Patterson, and Douglas O. Staiger. 2024. “The Impact and Implementation of
   Academic Interventions During COVID-19: Evidence from the Road to Recovery Project.”
   AERA Open 10.
Card, David and Alan B. Krueger. 1992. “Does School Quality Matter? Returns to Education and
   the Characteristics of Public Schools in the United States.” Journal of Political Economy
   100(1): 1-40.
Chetty, Raj, John N. Friedman, Nathaniel Hendren, Michael Stepner, and the Opportunity
   Insights Team. 2020. “How Did COVID-19 and Stabilization Policies Affect Spending and
   Employment? A New Real-Time Economic Tracker Based on Private Sector Data,” NBER
   Working Paper 27431.
Clemens, Jeffrey, and Anwita Mahajan. 2025. "Health Impacts of Federal Pandemic Aid to State
   and Local Governments." Available at SSRN: https://ssrn.com/abstract=5097211.
Clemens, Jeffrey, Philip G. Hoxie, John Kearns, and Stan Veuger. 2023. “How Did Federal Aid
   to State and Localities Affect Testing and Vaccine Delivery?” Journal of Public Economics
   225: 104972.
Clemens, Jeffrey, Philip G. Hoxie, and Stan Veuger. 2025. “Was Pandemic Fiscal Relief
   Effective Fiscal Stimulus? Evidence from Aid to State and Local Governments.” Journal of
   Macroeconomics 86: 103720.
Clemens, Jeffrey and Stan Veuger. 2021a. “Implications of the COVID-19 Pandemic for State
   Government Tax Revenues.” National Tax Journal 73(3): 619-644.
Clemens, Jeffrey and Stan Veuger. 2021b. “Politics and the Distribution of Federal Funds:
   Evidence from Federal Legislation in Response to COVID-19.” Journal of Public Economics
   204: 104554.
Congressional Research Service. 2009. “Funding for Education in the American Recovery and
   Reinvestment Act of 2009 (P.L. 111-5).” Congressional Research Service Report No.
   R40151.
Congressional Research Service. 2017. “History of the ESEA Title 1-A Formulas.”
   Congressional Research Service Report No. R44898.
Council of Economic Advisers. 2023. “Chronic Absenteeism and Disrupted Learning Require an
   All-Hands-on-Deck Approach.” The White House, September 13.
COVID-19 School Data Hub (CSDH). 2023a. “All State, School, and District COVID-19 Case
   Data.” Available at: https://www.covidschooldatahub.com/data-resources.
COVID-19 School Data Hub (CSDH). 2023b. “Community Case Rate Data By District.”
   Available at: https://www.covidschooldatahub.com/data-resources.
COVID-19 School Data Hub (CSDH). 2023c. “Percentage of School Year Spent In-Person,
   Hybrid, or Virtual.” Available at: https://www.covidschooldatahub.com/data-resources.
Dewey, Daniel C, Erin Fahle, Thomas J. Kane, Sean F. Reardon, and Douglas O. Staiger. 2024.
   “Federal Pandemic Relief and Academic Recovery.” NBER Working Paper 32897. Available
   at: https://www.nber.org/papers/w32897.

                                             31
Dewey, Daniel C, Erin Fahle, Thomas J. Kane, Sean F. Reardon, and Douglas O. Staiger. 2025.
    “Pivoting from Pandemic Recovery to Long-Term Reform: A District-Level Analysis.”
    Harvard University Center for Education Policy Research and The Educational Opportunity
    Project at Stanford University. Available at: https://prod-
    cepr.drupalsites.harvard.edu/sites/g/files/omnuum9881/files/cepr/files/pivoting_from_pande
    mic_recovery_to_long-term_reform_2.9.pdf.
Ed Data Express. 2024. Data Download Tool. US Department of Education. Available at:
    https://eddataexpress.ed.gov/download.
Fahle, Erin M, Thomas J. Kane, Tyler Patterson, Sean F. Reardon, Douglas O. Staiger, and
    Elizabeth A. Stuart. 2023. “School District and Community Factors Associated with
    Learning Loss During the COVID-19 Pandemic.” Center for Education Policy Research at
    Harvard University. Available at: https://www.notofman.com/wp-
    content/uploads/2024/03/explaining_covid_losses_5.23.pdf.
Fisher, Ronald. 1982. “Income and Grant Effects on Local Expenditures: The Flypaper Effect
    and Other Difficulties.” Journal of Urban Economics 12(3): 324-345.
Goldhaber, Dan, Thomas J. Kane, Andrew McEachin, Emily Morton, Tyler Patterson, and
    Douglas O. Staiger. 2023. “The Educational Consequences of Remote and Hybrid Instruction
    During the Pandemic.” American Economic Review: Insights 5(3): 377-92.
Goldhaber, Dan and Grace Falken. 2026. “ESSER and Student Achievement: Assessing the
    Impacts of the Largest One-Time Federal Investment in K12 Schools.” SSRN Working Paper
    No. 6269339. Available at: http://dx.doi.org/10.2139/ssrn.6269339.
Gordon, Nora. 2004. “Do Federal Grants Boost School Spending? Evidence from Title 1.”
    Journal of Public Economics 88(9): 1771-1792.
Gordon, Nora, and Sarah Reber. 2020. “Federal Aid to School Districts during the COVID-19
    Recession,” National Tax Journal 73(3): 781-804.
Gordon, Nora, and Sarah Reber. 2023. “Title I of ESEA: How the Formulas Work.” All4Ed.
    Available at: https://all4ed.org/publication/title-i-of-esea-how-the-formulas-work/.
Gramlich, Edward. 1977. “Intergovernmental Grants: A Review of the Empirical Literature.” In
    W. E. Oates (ed.), The Political Economy of Federalism, Lexington, MA: Lexington Books.
Green, Daniel, and Erik Loualiche. 2021. “State and Local Government Employment in the
    COVID-19 Crisis,” Journal of Public Economics 193: 104321.
Grembi, Veronica, Tommaso Nannicini, Ugo Troiano. 2016. “Do Fiscal Rules Matter?”
    American Economic Journal: Applied Economics 8(3):1-30.
Halloran, Clare, Claire E. Hug, Rebecca Jack, and Emily Oster. 2023. “Post COVID-19 Test
    Score Recovery: Initial Evidence from State Testing Data.” NBER Working Paper No.
    31113.
Halloran, Clare, Rebecca Jack, James C. Okun, and Emily Oster. 2021. “Pandemic Schooling
    Mode and Student Test Scores: Evidence from US States.” NBER Working Paper No. 29497.
Hanushek, Eric A. 1986. “Production and Efficiency in Public Schools,” Journal of Economic
    Literature 24(3): 1141-1177.
Hanushek, Eric A., Paul E. Peterson, Laura Talpey, and Ludger Woessmann. 2019. “The
    Unwavering SES Achievement Gap: Trends in U.S. Student Performance.” HKS Working
    Paper No. RWP19-012.

                                              32
Haughwout, Andrew, Benjamin Hyman, and Or Shachar. 2021. “The Option Value of Municipal
    Liquidity: Evidence from Federal Lending Cutoffs During Covid-19,” Mimeo: Federal
    Reserve Bank of New York.
Hines, James and Richard Thaler. 1995. “Anomalies: The Flypaper Effect.” Journal of Economic
    Perspectives 9(Fall): 217-321.
Hubbard, Glenn and Michael R. Strain. 2020. “Has the Paycheck Protection Program
    Succeeded?” Brookings Papers on Economic Activity, Fall: 335-390.
Institute of Education Sciences. 2024a. National Center for Education Statistics, Common Core
    of Data on America’s Public Schools. Available at: https://nces.ed.gov/ccd/files.asp.
Institute of Education Sciences. 2024b. Private School Universe Survey. National Center For
    Education Statistics. Available at: https://nces.ed.gov/surveys/pss/.
Jackson, C. Kirabo and Claire Mackevicius. 2021. “The Distribution of School Spending
    Impacts.” NBER Working Paper No. 28517.
Jackson, C. Kirabo, Rucker C. Johnson, and Claudia Persico. 2016. “The Effects of Spending on
    Educational and Economic Outcomes: Evidence from School Finance Reforms.” The
    Quarterly Journal of Economics 131(1): 157-218.
Korting, Christina, Carl Lieberman, Jordan Matsudaira, Zhuan Pei, and Yi Shen. 2023. “Visual
    Inference and Graphical Representation in Regression Discontinuity Designs.” The Quarterly
    Journal of Economics 138(3): 1977-2019.
Lake Geneva Schools. 2025. “Referendum Frequently Asked Questions.” Lake Geneva Schools.
    Available at: https://www.lakegenevaschools.com/about-us/referendum/referendum-
    frequently-asked-questions.
Lieberman, Mark. 2023. “Districts Proposed a Surge in Borrowing Last Year. Here’s How
    Voters Responded.” Education Week, June 8, 2023. Available at:
    https://www.edweek.org/leadership/districts-proposed-a-surge-in-borrowing-last-year-heres-
    how-voters-responded/2023/06.
Lutz, Byron. 2010. “Taxation with Representation: Intergovernmental Grants in a Plebiscite
    Democracy.” Review of Economics and Statistics 92(2): 316-332.
Missouri Census Data Center. 2024. Geocorr 2022. Available at:
    https://mcdc.missouri.edu/applications/geocorr.html.
MIT Election Data and Science Lab. 2024. "County Presidential Election Returns 2000-2020."
    Harvard Dataverse, V12. Available at: https://doi.org/10.7910/DVN/VOQCHQ.
National Council of State Legislatures. 2024. State Tax Actions Database. Accessible at:
    https://www.ncsl.org/fiscal/state-tax-actions-database.
New York Times. 2023. Coronavirus (COVID-19) Data in the United States. Available at
    https://github.com/nytimes/covid-19-data.
Office of Elementary and Secondary Education. 2020. “Frequently Asked Questions About the
    Elementary and Secondary School Emergency Relief Fund (ESSER Fund).” US Department
    of Education. Available at: https://www.ed.gov/media/document/frequently-asked-questions-
    esser-fund-7281.pdf.
Office of Elementary and Secondary Education. 2022. “Award Resources.” US Department of
    Education. Available at: https://oese.ed.gov/offices/office-state-grantee-relations-evidence-
    based-practices/state-and-grantee-relations/award-resources/.

                                               33
Office of Elementary and Secondary Education. 2023. “Elementary and Secondary School
   Emergency Relief Fund.” United States Department of Education. Available at:
   https://oese.ed.gov/offices/education-stabilization-fund/elementary-secondary-school-
   emergency-relief-fund/.
Oster, Emily, Rebecca Jack, Clare Halloran, John Schoof, Diana McLeod, Haisheng Yang, Julie
   Roche, and Dennis Roche. 2021. “Disparities in Learning Mode Access Among K–12
   Students during the COVID-19 Pandemic, by Race/Ethnicity, Geography, and Grade
   Level—United States, September 2020–April 2021.” Morbidity and Mortality Weekly
   Report 70 (26): 953.
Parolin, Zachary and Emma Lee. 2022. “US School Closure and Distance Learning Database.”
   OSF. Accessible at: https://osf.io/tpwqf/.
Peaster ISD. 2024. “PISD Calls for VATRE to Access Golden Pennies with No Tax Rate
   Increase Compared to Previous School Years.” Available at: https://www.peaster.net/vatre-
   2024#:~:text=The%20Board%20of%20Trustees%20met,Operations%20(M%26O)%20Rate
   %3A%20%240.6969.
Peeples, Julian. 2020. “CTA Locals Focus on Safety, Students, and Equity.” California Teachers
   Association. Available at: https://www.cta.org/educator/posts/bargaining-during-covid-19.
Reardon, S.F., E. M. Fahle, A. D. Ho, B.R. Shear, D. Kalogrides, J. Saliba, and T. J. Kane. 2024.
   Stanford Education Data Archive (Version SEDA 2023). Available at:
   https://edopportunity.org/get-the-data/seda-archive-downloads/#documentation-4-1.
Rivkin, S.G., Hanushek, E.A. and Kain, J.F., 2005. “Teachers, Schools, and Academic
   Achievement.” Econometrica 73(2): 417-458.
Schaffner, Brian, Marissa Shih, Stephen Ansolabeher, and Jeremy Pope. 2025. “Cooperative
   Election Study Common Content.” Harvard Dataverse, V9. Available at
   https://doi.org/10.7910/DVN/X11EP6.
Shoag, Daniel, Cody Tuttle, and Stan Veuger. 2019. “Rules Versus Home Rule: Local
   Government Responses to Negative Revenue Shocks.” National Tax Journal 72(3): 543-574.
Shores, Kenneth and Matthew P. Steinberg. 2022. “Has Federal Crisis Spending for K-12
   Schools Served its Intended Objectives?” The Brookings Institution, The Brown Center.
   Available at: https://www.brookings.edu/articles/has-federal-crisis-spending-for-k-12-
   schools-served-its-intended-objectives/.
Skinner, Rebecca R. 2016. “ESEA Title 1-A Formulas: In Brief.” Congressional Research
   Service.
Skinner, Rebecca R. and Cassandria Dortch. 2020. “CARES Act Education Stabilization Fund:
   Background and Analysis.” Congressional Research Service Report No. R46378.
Skinner, Rebecca R., Cassandria Dortch, and Joselynn H. Fountain. 2022. “Education
   Stabilization Fund Programs Funded by the CARES Act, CRRSAA, and ARPA: Background
   and Analysis.” Congressional Research Service Report No. R47027.
S&P Dow Jones Indices LLC. 2025. S&P CoreLogic Case-Shiller U.S. National Home Price
   Index [CSUSHPINSA]. Federal Reserve Bank of St. Louis. Available at:
   https://fred.stlouisfed.org/series/CSUSHPINSA.
UNESCO. 2023. “Global Monitoring of School Closures.” United Nations Education Scientific
   and Cultural Organization. Available at:

                                               34
   https://webarchive.unesco.org/web/20220629024039/https://en.unesco.org/covid19/education
   response/.
United States Bureau of Economic Analysis. 2024. Personal Consumption Expenditures: Chain-
   Type Price Index [PCEPI]. Retrieved from FRED, Federal Reserve Bank of Saint Louis.
   Available at: https://fred.stlouisfed.org/series/PCEPI.
United States Census Bureau. 2023. Inter-census County Population Estimates (2010-2019).
   United States Department of Commerce
United States Census Bureau. 2023. Small Area Income and Poverty Estimates (SAIPE)
   Program. United States Department of Commerce.
United States Department of Education. 2024. Education Stabilization Fund Public Transparency
   Portal API. Available at: https://covid-relief-data.ed.gov/api-documentation.
Urban Institute. 2024. Education Data Portal (Version 0.23.0). Urban Institute. Available at:
   https://educationdata.urban.org/data-explorer.
Wisconsin Department of Public Instruction. 2019. FY 2018-2019 Equalized Levy Rates (Mill
   Rates).
Wisconsin Department of Public Instruction. 2021. FY 2020-2021 Equalized Levy Rates (Mill
   Rates).
Wisconsin Policy Forum. 2021. “State Aid Slows Property Tax Growth.” Wisconsin Policy
   Forum Focus, December.




                                             35
Figures
Figure 1: Pandemic Learning Loss Compared to 2019 by District Poverty Share in 2022 and
2023 for Math and Reading Scores of 3rd Through 8th Graders

                                                 Panel A: Math




                                               Panel B: Reading




Note: This figure shows bins weighted by enrollments from SY 2020. The horizontal axis shows school district poverty
shares from the 2018 SAIPE data. The vertical axis shows changes in test scores from the spring of 2019, measured
in 2019 standard deviations using data from SEDA (Reardon, et al., 2024). Panel A shows changes from 2019 to 2022
and from 2019 to 2023 for math scores. Panel B shows changes from 2019 to 2022 and from 2019 to 2023 for reading
scores. All plots include fitted regression lines.




                                                        36
Figure 2: Discontinuities in Federal Revenues and ESSER Funds at the 5% Poverty Qualification
Threshold

 Panel A: Binned Federal Revenues in 2019                          Panel B: Un-Binned ESSER Funds Per
                 Per Pupil                                                        Pupil




  Panel C: Binned ESSER Funds Per Pupil                          Panel D: Event-Study of Federal Revenues
                                                                        and ESSER Funds Per Pupil




Note: This figure shows first stage scatter plots matching to Equations 4 and 3a in the text. Scatter plots in Panels A
and C use bins weighted by SY 2020 enrollment. Panel D shows coefficients from estimates of Equation 3a, described
in the text. We show the qualification threshold associated with the 5% poverty share with a vertical line.




                                                          37
Figure 3: Event Studies for the Impact of Qualifying for Additional ESSER Funds Per Pupil on
School District Revenues from Local Sources from SY 2017 to 2022




Note: This figure shows estimates of coefficient on the qualification dummy interactions from Equation 3b as
described in the text. These regressions correspond to Table 5 Column 4.




                                                    38
Figure 4: States That Made Any Increase or Decrease to their Property Taxes Before or After the
Start of the COVID-19 Pandemic




Note: This figure shows the share of states that made any change that increased their property tax levies (in blue) or
decreased their property tax levies (in red) for 2016-2019 and then 2020-2023. These data come from the National
Council of State Legislatures (2024).




                                                          39
Figure 5: Percent Changes in Electoral Engagement by Activity by ESSER Qualification Status
from 2016 to 2024




Note: This figure shows percent changes from 2018 levels for several measures of electoral engagement for
respondents in school districts within the difference-in-discontinuities sample. We use responses from the
Cooperative Election Study (Schaffner et al., 2025) to map political participation into school districts using
respondents zip codes and crosswalk files from (Missouri Census Data Center, 2024). The top row shows changes in
participation in the following activities for all respondents: voting, attending a local government meeting, and
contacting a public official. These changes are calculated as percentage changes from the 2018 baseline values for
districts above and below the 5% poverty share threshold for districts within our 50 formula child bandwidth. It is
important to not that district composition changes with the sample across survey years. We bootstrap 95%
confidence bands by resampling 500 times by zip code and qualification status. In the second row, we show the
parental share of respondents for each activity. This includes, the parental share of respondents who voted, the
parental share of respondents who attended a local meeting, and the parental share of respondents who contacted a
public official.




                                                        40
Tables

Table 1: Estimates of the Impact of Qualifying for Additional ESSER Funds on Measures of
School District Performance


                                              Math   Reading           Enrollment Absenteeism
             Qualification ∙ I{2023}         -0.0234 -0.00356             62.44
                                            (0.0213) (0.0214)           (78.61)
             Qualification ∙ I{2022}        0.00283 -0.0166               95.55      -381.1
                                            (0.0225) (0.0170)           (52.21)     (283.7)
             Qualification ∙ I{2021}                                     124.8*       120.2
                                                                        (58.96)     (164.7)
             Qualification ∙ I{2020}                                                  41.43
                                                                                    (81.40)
             First-Degree
             Polynomial                         Y             Y              Y                 Y
             LEA FEs                            Y             Y              Y                 Y
             Year FEs                           Y             Y              Y                 Y
             Observations                     6,547         5,962         32,732           25,910
             R2                               0.915         0.908          0.997             0.685
             Band                               50            50             50               50
             Combined Coefficient            -0.0206       -0.0202         282.8            -219.5
                                            (0.0413)      (0.0361)        (130.7)          (103.0)
             Pre-COVID Mean                   0.429         0.438         3,902.8            337.2
Note: This table shows estimates of Equation 3b as described in the text for math scores, reading scores, enrollment
counts, and absenteeism counts. Math scores and reading scores are measured in standard deviations of the 2019 test
score distribution. Enrollment and absenteeism are presented in student-years, where enrollment is measured in the
fall semester and absenteeism measures whether or not a student misses more than 10% of school days within a school
year. We present estimates for a bandwidth for 50 formula children from the 5% poverty threshold using a first-degree
polynomial specification. We present a linear combination of all post-COVID coefficients as well as the SY 2017 to
SY 2019 pre-COVID mean for all districts of the dependent variable at the bottom of the table. Regressions are
weighted by SY 2020 enrollments and standard errors clustered by district are shown in parentheses.

* p<0.05;** p<0.01; *** p<0.001




                                                         41
Table 2: Estimates of the Impact of Qualifying for Additional ESSER Funds on Enrollments by
Grade Level Groups
                                   Total      PK         K       1-5     6-8     9-12
                                    (1)       (2)       (3)      (4)     (5)       (6)
      Qualification ∙ I{2023}      62.44    -36.86    15.88     29.98  39.23*    9.246
                                  (78.61) (32.82) (12.35) (27.94) (19.15) (45.74)
      Qualification ∙ I{2022}      95.55    -32.10 22.39* 59.77* 33.96*          9.501
                                  (52.21) (34.37) (10.46) (24.12) (15.32) (31.78)
      Qualification ∙ I{2021}     124.8* 4.081        34.14     62.48  25.37*    8.281
                                  (58.96) (6.203) (21.80) (39.80) (10.40) (16.67)
      First-Degree Polynomial        Y         Y         Y        Y       Y         Y
      LEA FEs                        Y         Y         Y        Y       Y         Y
      Year FEs                       Y         Y         Y        Y       Y         Y
      Observations                32,732 22,010 30,800 31,192 30,870 22,939
      R2                           0.997    0.930     0.994     0.997   0.997    0.997
      Band                           50       50        50        50      50       50
      Combined Coefficient        282.81 -64.87       72.41    152.22   98.56    27.03
                                  (130.7) (66.9)      (43.3) (63.67) (42.00) (92.95)
      Pre-COVID Mean              3,902.8 98.73       277.6 1,450.0 943.1 1,490.7
Note: This table shows estimates of Equation 3b as described in the text for total enrollments, pre-kindergarten,
kindergarten, grades 1st through 5th, grades 6th through 8th, and grades 9th through 12th as counts. We present estimates
for 50 formula children from the 5% poverty threshold. We present a linear combination of all post-COVID
coefficients as well as the SY 2017 to SY 2019 pre-COVID mean for all districts of the dependent variable at the
bottom of the table. Regressions are weighted by SY 2020 enrollments and standard errors clustered by district are
shown in parentheses.

* p<0.05; ** p<0.01; *** p<0.001




                                                           42
Table 3: Estimates of the Impact of Qualifying for Additional ESSER Funds on Year-Over-Year
Cellphone Mobility Changes from CY 2019 Baseline
                                      Dep. Var: Mobility Change from 2019 Baseline
                                     2020                   2021                 2022
                                (1)         (2)        (3)         (4)      (5)         (6)
 Qualification Dummy        0.0530*** 0.0354 0.0897*** 0.0624* 0.108*** 0.0761*
                             (0.0146) (0.0215) (0.0212) (0.0294) (0.0262) (0.0384)
 First-Degree
 Polynomial                      Y                      Y                    Y
 Second-Degree
 Polynomial                                  Y                      Y                    Y
 Observations                  2,990       2,990      2,975       2,975    2,915      2,915
 R2                           0.0416      0.0430    0.0868       0.0885    0.100      0.101
 Band                           50           50        50           50      50          50
 Control Mean                 -0.307      -0.307     -0.263      -0.263   -0.209      -0.209
Note: This table shows estimates of Equation 4 as described in the text for year over year cellphone mobility changes
compared to 2019. We show first- and second-degree polynomial estimates using a bandwidth of 50 formula children
for 2020, 2021, and 2022. Below-cutoff means of the dependent variable are presented at the bottom of the table. We
show robust standard errors in parentheses. We weight estimates by SY 2020 enrollments.

* p<0.05;** p<0.01; *** p<0.001




                                                         43
Table 4: Estimates of the Impact of Qualifying for Additional ESSER Funds on School District
Staffing Per 1000 Pupils
                              Tot.                 Non-                   Inst.     Dist.
                             Staff   Teachers Teachers Counselors        Aids     Admin
                              (1)        (2)        (3)          (4)       (5)       (6)
Qualification ∙ I{2023}      0.160     -0.316      0.477       -0.011    1.231   0.722***
                            (4.487)   (0.984)     (4.331)     (0.132)  (1.028)    (0.179)
Qualification ∙ I{2022}      3.089      0.968      2.121       0.101     0.666     0.050
                            (1.793)   (0.595)     (1.597)     (0.089)  (0.600)    (0.079)
Qualification ∙ I{2021}      0.616      0.203      0.414       -0.095    0.166    0.0043
                            (1.574)   (0.571)     (1.243)     (0.075)  (0.584)    (0.062)
First-Degree
Polynomial                     Y          Y          Y            Y         Y         Y
LEA FEs                        Y          Y          Y            Y         Y         Y
Year FEs                       Y          Y          Y            Y         Y         Y
Observations                30,247    30,247      30,247      29,383    29,154    28,973
R2                           0.768      0.778      0.771       0.748     0.849     0.812
Band                           50        50          50          50        50         50
Combined Coefficient         3.866      0.854      3.012       -0.005    2.063       0.777
                            (6.205)   (1.691)     (5.743)     (0.248)  (1.813)      (0.255)
Pre-COVID Mean               146.9      71.34      75.52       2.233     19.37       2.172
Note: This table shows estimates of Equation 3b as described in the text for total staff per 1000 pupils, teachers per
1000 pupils, non-teaching staff per 1000 pupils, guidance counselors per 1000 pupils, instructional aides per 1000
pupils, and district administrators per 1000 pupils. Per pupil measures use SY 2020 enrollments. We present estimates
for 50 formula children from the 5% poverty threshold and a first-degree polynomial specification. We present a linear
combination of all post-COVID coefficients as well as the SY 2017 to SY 2019 pre-COVID mean for all districts of
the dependent variable at the bottom of the table. Regressions are weighted by SY 2020 enrollments and standard
errors clustered by district are shown in parentheses.

* p<0.05;** p<0.01; *** p<0.001




                                                         44
 Table 5: Estimates of the Impact of Qualifying for Additional ESSER Funds on School District
 Revenues and Expenditures Per Pupil

                                                         Panel A: Revenues
                                                    Sub-Totals                        Exhibits
                                     Total     Fed.    State       Local     Prop. Tax.     St. Fmla.
                                      (1)       (2)     (3)          (4)          (5)           (6)
Qualification ∙ I{2022}             -357.9    89.44    52.17     -499.5**     -390.8**         54.30
                                   (214.3)  (50.88)  (122.3)      (168.8)      (125.0)       (65.62)
Qualification ∙ I{2021}             -182.1  148.8**   -29.94      -300.9*    -382.8***         117.6
                                   (245.5)  (45.80)  (207.5)      (146.8)      (115.9)       (63.44)
Qualification ∙ I{2020}             -3.504    23.16    80.12       -106.8        55.14         15.39
                                   (173.5)  (33.03)  (82.42)      (148.4)      (85.58)       (44.38)
First-Degree Polynomial                Y         Y       Y            Y            Y             Y
LEA FEs                                Y         Y       Y            Y            Y             Y
Year FEs                               Y         Y       Y            Y            Y             Y
Observations                       28,056    28,056  28,056       28,056       26,021        28,056
R2                                   0.898    0.862    0.820       0.940         0.955         0.913
Band                                  50        50      50           50           50            50
Combined Coefficient                -543.5    261.3  102.36        -907.2       -718.4         187.3
                                   (486.1)  (108.0)  (308.6)      (373.9)      (243.5)       (138.6)
Pre-COVID Mean                    18,015.1    589.0  6,282.6     11,143.4      8,726.1       3,908.6
                                                       Panel B: Expenditures
                                     Total  ELSEC. Capital       Instruct.     Salaries      Benefits
                                      (1)       (2)     (3)          (4)          (5)           (6)
Qualification ∙ I{2022}              192.3    85.67    7.600       87.54         102.3        -48.80
                                   (555.6)  (138.0)  (505.8)      (87.46)      (72.04)       (61.17)
Qualification ∙ I{2021}              454.0    64.00    266.4       61.75         83.97        -63.07
                                   (463.4)  (208.7)  (406.1)      (130.9)      (53.54)       (191.9)
Qualification ∙ I{2020}              99.22    32.61   -92.44       41.99        -3.243         70.90
                                   (369.3)  (95.81)  (310.0)      (56.82)      (51.84)       (40.04)
First-Degree Polynomial                Y         Y       Y            Y            Y             Y
LEA FEs                                Y         Y       Y            Y            Y             Y
Year FEs                               Y         Y       Y            Y            Y             Y
Observations                       28,056    28,056  28,056       28,056       28,056        28,056
R2                                   0.829    0.956    0.346       0.956         0.973         0.929
Band                                  50        50      50           50           50            50
Combined Coefficient                 745.5    182.3    181.6       191.3         183.0         -41.0
                                  (1,217.2) (317.7) (1,112.5)     (196.1)      (147.1)       (214.2)
Pre-COVID Mean                    17,992.8 14,887.4 1,752.8       9,125.8      8,438.2       3,859.4
 Note: This table shows estimates of Equation 3b as described in the text for revenues in Panel A and expenditures in
 Panel B. In Panel A we show estimates for total revenues per pupil, federal revenues per pupil, state revenues per
 pupil, local (own source) revenues per pupil, property tax revenues per pupil and state formula revenues per pupil. In
 Panel B, we show total expenditures per pupil, elementary and secondary education expenditures per pupil, capital
 outlays per pupil, expenditures on instruction (including compensation) per pupil, salaries for all employees per pupil,

                                                           45
and benefits for all employees per pupil. Per pupil measures use SY 2020 enrollments. We present estimates for 50
formula children from the 5% poverty threshold and a first-degree polynomial specification. We present a linear
combination of all post-COVID coefficients as well as the SY 2017 to SY 2019 pre-COVID mean for all districts of
the dependent variable at the bottom of the table. Regressions are weighted by SY 2020 enrollments and standard
errors clustered by district are shown in parentheses.

* p<0.05;** p<0.01; *** p<0.001




                                                       46
Appendix Figures
Appendix Figure 1: National Shares of School District Revenues and Expenditures

                                               Panel A: Revenues




                                             Panel B: Expenditures




Note: This figure shows national shares of revenues by source, federal, state, and local in panel A over time. In Panel
B, the figure shows shares of total expenditures for instruction, support services, and capital outlays over time. Data
for this figure comes from IES (2024).




                                                          47
Appendix Figure 2: Map of Districts Included in the 50 Formula Child Bandwidth Difference-in-
Discontinuities Sample




Note: This figure shows districts in the 50 formula child bandwidth sample in blue and all other districts in red.




                                                          48
Appendix Figure 3: Covariate Smoothness at the 5% Poverty Qualification Cutoff Before
COVID in Trends and Levels
                                                Panel A: Trends




                                                Panel B: Levels




Note: This figure shows balance checks for covariate differences across the qualification cutoff as pre-COVID trends
in Panel A and in levels in Panel B akin to equation 4 in the text. Scatters are weighted by SY 2020 enrollment. We
show the qualification threshold associated with the 5% poverty share with a vertical dashed line. We show trends
from SY 2017 to SY 2020 for enrollment, White share, Black share, staff per pupil, and teachers per pupil and trends
from SY 2017 to SY 2019 for absenteeism, federal revenues per pupil, total revenues per pupil, reading scores, and
math scores.



                                                        49
Appendix Figure 4: National Enrollment Trends in Elementary and Secondary Schools
                                            Panel A: Total Enrollments




                           Panel B: Indexed Enrollment Trends by Grade Group




Note: Panel A of this figure shows national enrollment trends for all public schools. Panel B shows indexed enrollment
trends for public schools for grades kindergarten through 5 th (blue line), 6th through 8th (red line), and 9th through 12th
(green line). This figure uses data from the Common Core of Data (IES, 2024b).




                                                            50
Appendix Figure 5: Event Studies for the Impact of Qualifying for Additional ESSER Funds Per
Pupil on School District Performance from SY 2017 to 2023
                 Panel A: Math                                            Panel B: Reading




             Panel C: Enrollment                                       Panel D: Absenteeism




Note: These figures show coefficients from the qualification dummy interactions for each year as described in
Equation 3b. These regressions correspond to Columns 2, 5, 8, and 11 of Table 1.




                                                     51
Appendix Figure 6: Binned Scatter of Changes in Enrollment from SY 2020 Across the 5%
Poverty Qualification Cutoff in SY 2021, 2022, and 2023




Note: This figure shows weighted binned scatters for changes in enrollment from SY 2020 on the vertical axis and the
centered formula child count running variable on the horizontal axis. We show the qualification threshold associated
with the 5% poverty share with a vertical dashed line.




                                                        52
Appendix Figure 7: Binned Scatter of Year Over Year Changes in Cell Phone Mobility from
2019 Across the 5% Poverty Qualification Cutoff in CY 2020, 2021, and 2022




Note: This figure shows weighted binned scatters for changes in year over year cell phone mobility around schools
on the vertical axis and the centered formula child count running variable on the horizontal axis. We show the
qualification threshold associated with the 5% poverty share with a vertical dashed line.




                                                        53
Appendix Tables
Appendix Table 1: Distract Financial Snapshot Around the 5% Poverty Share Qualification
Cutoff in SY 2019

                                                    Below                   Above          Total
                                                      (1)                     (2)           (3)
                  Total Revenues                   19,511.7                17,816.8      18,407.9
                                                   (7,085.5)               (7,560.9)     (7,441.0)
                  Federal Revenues                   455.3                   680.6         602.0
                                                    (527.3)                (1034.8)       (897.6)
                  State Revenues                    5,764.8                 6,773.2       6,421.5
                                                   (2,500.8)               (3,795.3)     (3,433.6)
                  Local Revenues                   13,291.6                10,363.0      11,384.5
                                                   (7,169.3)               (7,328.3)     (7,404.4)
                  Total Expenditures               19,714.9                17,827.6      18,485.9
                                                   (7,092.3)               (7,590.7)     (7,473.5)
                  Instruction Current Expenditures 10,039.7                 8,839.7       9,258.2
                                                   (4,039.1)               (3,429.9)     (3,697.4)
                  Salaries for Instruction          6,397.2                 5,553.9       5,848.0
                                                   (2,592.9)               (2,187.1)     (2,370.3)
                  Enrollment                        5,235.0                 3,261.9       3,950.1
                                                   (7,262.3)               (4,956.6)     (5,937.7)
                  N                                   693                    3,983         4,676
Note: This table shows means and standard deviations for select variables for the sample of districts in the 50 formula
child bandwidth sample. All revenue and expenditure variables are on a per pupil basis using SY 2020 enrollments.
Instruction current expenditures includes salaries for instruction. Column 1 shows districts below the 5% poverty
threshold, Column 2 shows districts above the threshold, and Column 3 shows all districts in the 50 formula child
bandwidth sample. We weight calculations by SY 2020 enrollments.




                                                          54
Appendix Table 2: Consolidated Timeline Relevant for the Delivery of ESSER Funds

      Event                                                                             Calendar Date
      Every Student Succeeds Act Signed into Law                                             2015
      Every Student Succeeds Act Fully Implemented                                           2017
      SAIPE Data Collection for FY2020 Title 1-A Funds                                       2018
      Title 1-A Appropriation Levels Set by Congress for FY2020                              2019
      First US COVID Case Confirmed by CDC                                               January 2020
      First COVID School Closure in Washington State                                    February 2020
      Families First Coronavirus Response Act Signed into Law                             March 2020
      CARES Act Signed into Law, Appropriating $30.75B for ESSER I                        March 2020
      CRRSA Signed into Law, Appropriating $54.3B for ESSER II                          December 2020
      ARP Signed into Law, Appropriating $122B for ESSER III                              March 2021
      First US COVID-19 Vaccine Delivered to the Public                                 December 2021
      Spending Deadline for ESSER I                                                       October 2022
      Spending Deadline for ESSER II                                                      October 2023
      Spending Deadline for ARP ESSER (ESSER III)                                         October 2024
Note: This table shows a timeline of selected events relevant for COVID-19 and the delivery of ESSER funds.




                                                       55
       Appendix Table 3: Select Difference-in-Discontinuities Sample Summary Statistics Across
       Sample School Years

                          2017      2018      2019       2020      2021      2022    2023      Total
                           (1)       (2)       (3)        (4)       (5)       (6)     (7)       (8)
Math Scores               0.428     0.429     0.428                          0.306   0.356     0.390
                         (0.267)   (0.268)   (0.272)                       (0.278)  (0.297)  (0.280)
Reading Scores            0.470     0.439     0.407                          0.338   0.399     0.411
                         (0.255)   (0.245)   (0.245)                       (0.264)  (0.287)  (0.262)
Enrollment               3858.3    3,897.0   3,910.2   3,950.1   3,863.3   3,898.3  3,920.6  3,899.7
                        (5675.9) (5,758.8) (5,824.5) (5937.7) (5,783.1) (5,896.6) (5,944.1) (5,831.8)
Total Schools             6.209     6.227     6.221      6.277     6.278     6.379   6.412     6.286
                         (6.592)   (6.679)   (6.767)   (6.935)   (6.912)   (7.118)  (7.364)  (6.914)
Total Revenues          17,611.2 17,937.2 18,466.1 18,628.8 19,110.8 19,312.5                18511.1
                        (6788.4) (6,908.6) (7,463.1) (8,234.6) (7,122.4) (15,158.6)         (9127.8)
Federal Revenues          577.1     586.4     604.3      604.0   1,049.6   1,435.7             809.5
                         (694.4)   (849.6)   (918.7)   (727.4)   (988.9)  (2,003.4)         (1169.4)
Fed. Rev. w/ ESSER        577.5     585.8     602.0      598.3   2,206.4   1,046.9             936.1
                         (709.4)   (848.3)   (897.6)   (723.5) (1,797.0)   (856.5)          (1198.4)
Teachers per Pupil       0.0706    0.0711    0.0714    0.0718    0.0739     0.0744  0.0751    0.0726
                        (0.0190) (0.0191) (0.0192) (0.0191) (0.0196) (0.0199) (0.0201) (0.0195)
Enr. Share White          0.785     0.777     0.770      0.763     0.756     0.746             0.766
                         (0.166)   (0.169)   (0.173)   (0.176)   (0.177)   (0.180)           (0.174)
Mobility Change                                         -0.266    -0.179   -0.0973            -0.181
from 2019                                              (0.155)   (0.225)   (0.277)           (0.235)
Formula Children          233.1     218.5     212.5      197.2     191.8     238.5             215.3
                         (302.3)   (288.1)   (299.0)   (267.0)   (279.7)   (350.1)           (299.3)
Pop Ages 5 to 17         4,074.2   4,071.7   4,061.4   4,056.8   4,063.8   4,350.1           4,113.0
                        (5,924.4) (5,969.7) (6,002.7) (6,051.0) (6,122.4) (6,548.8)         (6,107.1)
       Note: This table shows summary statistics for the 50 formula child bandwidth sample for each sample year, as data
       are available. Means are presented with standard deviations in parentheses. We weight calculations by SY 2020
       enrollments. Math and reading scores are presented in standard deviations from the 2019 testing distribution.
       Enrollment, total schools, formula children, and population ages 5 to 17 are presented in counts. Total revenues,
       federal revenues, and federal revenues with ESSER funds are presented in 2023 USD per enrollee in SY 2020.
       Teachers per pupil (PP) are full time equivalent teacher counts per enrollee in SY 2020. Enrollment share white is
       presented as a share, which ranges from 0 to 1, and mobility change from 2019 is presented as a share, where -0.2
       reflects a 20% decline in mobility from 2019.




                                                              56
Appendix Table 4: Variable Descriptions
 Variable             Description                           Measurement period      Source
 ESSER Funds Per      Real funds from ESSER I, II, and      2020-2021               DOE
 Pupil                III per Fall 2019 enrollee. Top                               Transparency
                      0.01% of observations have been                               API
                      trimmed
 ESSER Funds Per      ESSER Funds, which are added          End of school year.     CCD, DOE
 Pupil Series (for    as a lump sum in SY 2021,                                     API
 event study first    appended to the federal revenue
 stage)               series with reported ESSER funds
                      from the special exhibits
                      removed.
 Qualification        Takes a value of 1 if a district in   2018                    SAIPE
 Dummy                SY 2019 has a poverty share
                      above 5%, which is calculated
                      using data from the 2018 SAIPE
                      estimates of kids in poverty and
                      all kids in the district area (not
                      necessarily enrolled). Kids in
                      poverty are formula kids.
 Centered Formula     The number of formula children        2018                    SAIPE
 Child Count          in a district (from SAIPE)
                      subtracted from that district's
                      normalized 5% poverty threshold.
                      The threshold is the number of
                      formula kids that would constitute
                      a 5% share of their total district
                      school age pop. These are
                      measured in 2017.
 Enrollment (and      Counts of children enrolled (not      Fall of each school     CCD
 detailed enrollment) the same as school aged pop)          year
 Absenteeism          Counts of students determined to      End of school year      DOE
                      be chronically absent (DOE
                      definition)
 Teachers PP          Full time equivalent teacher count    Fall of each school     CCD
                      per enrollee (both current)           year
 Total Staff PP       Full time equivalent staff count      Fall of each school     CCD
                      per enrollee (both current)           year
 Spending PP Vars     Real dollars per enrollee (both       End of school year      CCD
                      current values)
 Covid Cases from     Area Covid Case Rate per 100k,        Throughout the school   Covid
 Area (Zip)           attached by district HQ zip code      year                    School Data
                                                                                    Hub (Oster)
 Movement Changes Cell phone movement change                Throughout the          Parolin and
                  compared to same period in 2019.          calendar year           Lee (2021)

                                              57
Fixed Spending PP    Real dollars per enrollee in SY       End of school year       CCD
Vars                 2020
White Share          White fraction of students.           Beginning of school      CCD
                                                           year
Black Share          Black fraction of students.           Beginning of school      CCD
                                                           year
Federal Revenues     All federal revenues, including       End of the school year   CCD
PP                   emergency funds and regular
                     transfers such as Title 1, for each
                     FY. These are divided by the
                     number of pupils in SY 2020
                     (Fall 2019).
Revenues PP          All revenues for each FY. These       End of the school year   CCD
                     are divided by the number of
                     pupils in SY 2020 (Fall 2019).
Reading Scores       Average NAEP score for Reading        Spring of every other    SEDA
                     for the entire district for 3-8th     school year: 2017,
                     grades, scaled in 2019 standard       2019, 2022, 2023
                     deviations and demeaned.              (next would be 2025).
Math Scores          Average NAEP score for Math           Spring of every other    SEDA
                     for the entire district for 3-8th     school year: 2017,
                     grades, scaled in 2019 standard       2019, 2022, 2023
                     deviations and demeaned.              (next would be 2025).
State Revenues PP    All state revenues, including         End of the school year   CCD
                     formula assistance and other state
                     funds, divided by pupils in SY
                     2020 (Fall 2019). These do not
                     include funds passed through the
                     state, such as Title 1.
State Formula        State revenues from funding           End of the school year   CCD
Funds PP             formulas (e.g. local control
                     funding in CA) divided by pupil
                     in SY 2020 (Fall 2019).
Local Revenues       All local revenues, including         End of the school year   CCD
                     from parcel, sales, property, and
                     other taxes and revenue sources,
                     divided by pupils in SY 2020
                     (Fall 2019)
Property Tax         Local revenues from property          End of the school year   CCD
Revenues             taxes divided by pupils in SY
                     2020 (Fall 2019).
Total Expenditures   Total current expenditures            End of the school year   CCD
                     divided by pupils in SY 2020
                     (Fall 2019).




                                             58
Current               Total expenditures on K-12        End of the school year        CCD
Expenditures on       education divided by pupils in SY
Elementary and        2020 (Fall 2019).
Secondary
Education
Capital Outlays       Total capital outlays for all types    End of the school year   CCD
                      of capital, including buildings,
                      divided by pupils in SY 2020
                      (Fall 2019).
Instruction Current   Expenditures on classroom              End of the school year   CCD
Expenditures          instruction, including salaries and
                      supplies, divided by pupils in SY
                      2020 (Fall 2019).
Salaries              Expenditures on salaries for all       End of the school year   CCD
                      employees divided by pupils in
                      SY 2020 (Fall 2019).
Benefits              Expenditures on Benefits for all       End of the school year   CCD
                      employees divided by pupils in
                      SY 2020 (Fall 2019).
Property Tax          State property tax changes. For        Throughout the school    NCSL
Increases or          increases, states that have any        year
Decreases             action that increases property
                      taxes gets coded as a 1. For
                      Decreases, states that have any
                      action that decreases property
                      taxes gets coded as a 1. These are
                      not mutually exclusive.
School Staff PP       Count of full-time equivalent staff    Fall of each school      CCD
                      that work at the schools in district   year
                      divided by pupils in SY 2020.
Guidance              Count of full-time equivalent          Fall of each school      CCD
Counselors PP         guidance counselors divided by         year
                      pupils in SY 2020.
Instructional Aides   Count of full time equivalent          Fall of each school      CCD
PP                    instructional aides divided by         year
                      pupils in SY 2020.
District Admin per    Count of full-time equivalent          Fall of each school      CCD
pupil                 district administrative staff          year
                      divided by pupils in SY 2020.




                                               59
Appendix Table 5: Comparison of All US School Districts to the Difference-in-Discontinuities
Sample With a 50 Formula Children Bandwidth
                                                                  Diff-in-       Out of Diff-
                                                     All           Disc             in-Disc
                                                  Districts       Sample            Sample
                                                     (1)            (2)                (3)
                    Math Scores                    0.0132          0.429            -0.0143
                                                   (0.363)        (0.269)           (0.351)
                    Reading Scores                 0.0428          0.438             0.0167
                                                   (0.331)        (0.250)           (0.318)
                    Enrollment                    43,456.8        3,888.5          47,082.9
                                                 (86,565.3)      (5,753.0)        (89,558.5)
                    Total Schools                   62.17          6.219              67.29
                                                   (128.6)        (6.679)           (133.1)
                    Total Revenues                14,544.4       18,004.8          14,227.3
                                                  (4,914.2)      (7,067.8)         (4,537.4)
                    Federal Revenues               1,070.4         589.3            1,114.5
                                                   (726.1)        (826.2)           (699.9)
                    Fed. Rev. with ESSER           1,085.8         588.4            1,131.4
                                                   (753.9)        (822.3)           (730.5)
                    Teachers PP                    0.0618         0.0710             0.0610
                                                  (0.0142)       (0.0191)          (0.0133)
                    Enr. Share White                0.493          0.778              0.467
                                                   (0.288)        (0.169)           (0.282)
                    Formula Children               9,502.9         221.4           10,353.5
                                                 (23,317.3)       (296.6)         (24,184.8)
                    Pop Ages 5 to 17              49,306.7        4,069.1          53,452.3
                                                (100,347.1)      (5,965.3)       (103,848.4)
Note: This table shows summary statistics for select variables measured pre-COVID as averages between SY 2017
and SY 2019. Column 1 shows all public school districts, Column 2 shows districts within the 50 formula child
bandwidth sample, and Column 3 shows districts not in the 50 formula child bandwidth sample. Means are presented
with standard deviations in parentheses. We weight calculations by SY 2020 enrollments. Math and reading scores
are presented in standard deviations from the 2019 testing distribution. Enrollment, total schools, formula children,
and population ages 5 to 17 are presented in counts. Total revenues, federal revenues, and federal revenues with
ESSER funds are presented in 2023 USD per enrollee in SY 2020. Teachers per pupil (PP) are full time equivalent
teacher counts per enrollee in SY 2020. Enrollment share white is presented as a share, which ranges from 0 to 1.




                                                         60
Appendix Table 6: First Stage Regression Discontinuity Estimates of ESSER Funding at the 5%
Poverty Qualification Cutoff

                                               45                 50                 55
                                         (1)        (2)     (3)        (4)     (5)        (6)
 Qualification Dummy                  390.0*** 467.1*** 388.2*** 452.4*** 358.6*** 474.7***
                                       (71.57)    (112.7) (66.27)    (109.0) (62.99)    (105.8)
 First-Degree Polynomial                  Y                  Y                  Y
 Second-Degree Polynomial                            Y                  Y                  Y
 Observations                           4,413      4,413   4,696      4,696   4,986      4,986
 R2                                    0.0605     0.0606  0.0669     0.0670  0.0770     0.0775
 F-Statistic                            29.69      17.17   34.31      17.24   32.41      20.13
 Below Mean                             816.9      816.9   808.4      808.4   798.6      798.6
Note: This table shows estimates of Equation 4 as described in the text. F-statistics for the qualification dummy and
below-cutoff means are presented at the bottom of the table. We show robust standard errors in parentheses. We weight
estimates by SY 2020 enrollments.

* p<0.05;** p<0.01; *** p<0.001




                                                         61
62
Appendix Table 7: Test for Pre-Treatment Differences at the 5% Poverty Qualification Cutoff Before COVID in Trends
                                                                 Panel A: First-Degree Polynomial
                                           Absent-          Tot.      Fed.      Teachers                             Enroll-         Share      Share
               Math   Reading               eeism           Rev       Rev          PP        Staff PP                  ment          White      Black
 Band           (1)      (2)                 (3)            (4)        (5)         (6)          (7)                     (8)            (9)       (10)
 45          0.00491 0.0122                 31.65          267.5     6.494 -0.0000453 -0.00165                        -146.1        0.0105*   -0.00230
             (0.0180) (0.0165)             (23.64)        (253.8) (38.06) (0.000667) (0.00186)                       (171.1)       (0.00435) (0.00121)
 50          -0.00462 0.00702               18.51          213.4 0.0943 -0.000841            -0.00134                 -137.8       0.00941* -0.00181
             (0.0170) (0.0161)             (21.49)        (226.2) (35.79) (0.000642) (0.00166)                       (167.1)       (0.00413) (0.00102)
 55          -0.00134 0.00894               15.73          241.6 -2.493         -0.00103    -0.000866                 -132.2       0.00896* -0.00153
             (0.0164) (0.0154)             (20.16)        (208.4) (33.90) (0.000624) (0.00156)                       (162.9)       (0.00401) (0.000956)
                                                               Panel B: Second-Degree Polynomial
                                           Absent-          Tot.      Fed.      Teachers                             Enroll-         Share           Share
               Math        Reading          eeism           Rev       Rev          PP        Staff PP                  ment          White           Black
 Band           (1)           (2)            (3)            (4)        (5)         (6)          (7)                     (8)            (9)            (10)
 45           0.0166       0.0455*          44.19          161.7     1.247 0.0000974        0.000876                  63.41         0.00549       0.000450
             (0.0251)      (0.0221)        (43.34)        (395.8) (47.57) (0.000911) (0.00247)                       (177.4)       (0.00457)      (0.00142)
 50           0.0241       0.0426*          53.06          253.9     10.49      0.000976 -0.0000628                   -3.560        0.00815       -0.000877
             (0.0237)      (0.0207)        (40.54)        (344.3) (47.22) (0.000910) (0.00238)                       (173.4)       (0.00458)      (0.00120)
 55           0.0128        0.0311          49.24          191.3     12.31      0.000691    -0.000994                 -39.97        0.00885        -0.00134
             (0.0232)      (0.0198)        (38.81)        (325.2) (46.95) (0.000878) (0.00237)                       (170.4)       (0.00462)      (0.00127)
Note: Note: This table shows estimates for the coefficient of the qualification dummy in Equation 4 as described in the text for covariate pre-COVID trends. These
coefficients are shown for the 45, 50, and 55 formula child bandwidth samples as well as first-degree polynomial estimates in Panel A and second-degree polynomial
estimates in Panel B. We show robust standard errors in parentheses. We weight estimates by SY 2020 enrollments. We show trends from SY 2017 to SY 2020 for
enrollment, White share, Black share, staff per pupil, and teachers per pupil and trends from SY 2017 to SY 2019 for absenteeism, federal revenues per pupil, total
revenues per pupil, reading scores, and math scores.

 * p<0.05;** p<0.01; *** p<0.001




                                                                                63
Appendix Table 8: Test for Covariate Smoothness in Levels at the 5% Poverty Qualification Cutoff Before COVID in Levels
                                                              Panel A: First-Degree Polynomial
                                         Absent-          Tot.                 Teachers                              Enroll-        Share     Share
             Math   Reading               eeism           Rev     Fed. Rev         PP      Staff PP                   ment          White     Black
 Band         (1)      (2)                  (3)           (4)        (5)           (6)        (7)                      (8)            (9)      (10)
 45         -0.0480 -0.0151               -284.0         787.4 154.8*** 0.00714*** 0.0115*                          -4170.7*       0.0592* -0.00519
           (0.0392) (0.0389)             (151.9)        (879.4)    (46.64)     (0.00212) (0.00525)                  (2007.7)       (0.0258) (0.00571)
 50         -0.0430 -0.0177               -207.1         160.1     134.8** 0.00544** 0.00941                         -3532.0       0.0535* -0.00273
           (0.0367) (0.0364)             (153.5)        (838.9)    (42.80)     (0.00204) (0.00501)                  (1996.3)       (0.0245) (0.00528)
 55         -0.0305 -0.00686              -169.1         349.0     126.4** 0.00550** 0.00816                         -3223.5       0.0502* -0.00299
           (0.0354) (0.0350)             (155.5)        (808.8)    (40.03)     (0.00196) (0.00485)                  (1975.2)       (0.0238) (0.00517)
                                                            Panel B: Second-Degree Polynomial
                                         Absent-          Tot.                 Teachers                              Enroll-        Share          Share
             Math        Reading          eeism           Rev     Fed. Rev         PP      Staff PP                   ment          White          Black
 Band         (1)           (2)             (3)           (4)        (5)           (6)        (7)                      (8)            (9)           (10)
 45         -0.0437       -0.0205         -213.0        1440.9      76.63     0.00857**     0.0137                   -1709.0        0.0259      0.000519
           (0.0621)      (0.0636)        (153.1)       (1229.5) (80.51)        (0.00316) (0.00787)                  (1224.2)       (0.0306)     (0.00821)
 50         -0.0515       -0.0171        -328.0*        1999.5      123.3     0.0102*** 0.0157*                     -3154.9*        0.0417       -0.00387
           (0.0578)      (0.0592)        (150.1)       (1158.3) (76.36)        (0.00296) (0.00740)                  (1328.7)       (0.0285)     (0.00754)
 55         -0.0653       -0.0311        -338.9*        1344.2      134.8     0.00901** 0.0160*                     -3540.0*        0.0490       -0.00289
           (0.0554)      (0.0563)        (142.1)       (1127.4) (72.23)        (0.00284) (0.00707)                  (1421.2)       (0.0282)     (0.00728)
Note: Note: This table shows estimates for the coefficient of the qualification dummy in Equation 4 as described in the text for covariates in levels. These coefficients
are shown for the 45, 50, and 55 formula child bandwidth samples as well as first-degree polynomial estimates in Panel A and second-degree polynomial estimates
in Panel B. We show robust standard errors in parentheses. We weight estimates by SY 2020 enrollments.

 * p<0.05;** p<0.01; *** p<0.001




                                                                                   64
        Appendix Table 9: Estimates of the Impact of Qualifying for Additional ESSER Funds on Measures of School District Performance
                                             Math                      Reading                                       Enrollment                          Absenteeism
                                 (1)          (2)      (3)      (4)      (5)                        (6)         (7)     (8)      (9)                (10)    (11)     (12)
Qualification ∙ I{2023}       -0.0389       -0.0234 -0.0233 -0.0227 -0.00356                    -0.00459       89.22   62.44    28.90
                             (0.0221)      (0.0213) (0.0205) (0.0224) (0.0214)                  (0.0202)      (80.85) (78.61) (77.12)
Qualification ∙ I{2022}      -0.00305      0.00283 0.00499 -0.0322 -0.0166                       -0.0131      119.5*   95.55    70.16              -457.9      -381.1        -369.4
                             (0.0234)      (0.0225) (0.0217) (0.0176) (0.0170)                  (0.0162)      (54.92) (52.21) (50.67)             (283.7)     (283.7)       (281.3)
Qualification ∙ I{2021}                                                                                       150.5* 124.8*     99.58              183.3       120.2         103.3
                                                                                                              (60.46) (58.96) (58.33)             (163.7)     (164.7)       (165.9)
Qualification ∙ I{2020}                                                                                                                            51.87       41.43         42.33
                                                                                                                                                  (78.29)     (81.40)       (84.41)
First-Degree
Polynomial                       Y             Y        Y        Y                    Y             Y            Y       Y               Y           Y           Y             Y
LEA FEs                          Y             Y        Y        Y                    Y             Y            Y       Y               Y           Y           Y             Y
Year FEs                         Y             Y        Y        Y                    Y             Y            Y       Y               Y           Y           Y             Y
Observations                   5,991         6,547   7,036    5,427                 5,962         6,459       3,0751 3,2732           3,4762      2,4311      2,5910        2,7521
R2                             0.914         0.915   0.916    0.907                 0.908         0.912        0.997   0.997           0.997       0.685       0.685         0.684
Band                             45            50      55       45                    50            55           45      50              55          45          50            55
Combined Coefficient          -0.0419       -0.0206 -0.0183 -0.0550                -0.0202       -0.0177       359.1   282.8           198.6       -222.7      -219.5        -223.8
                             (0.0428)      (0.0413) (0.0397) (0.0375)             (0.0361)      (0.0341)      (138.2) (130.7)         (127.8)     (91.27)     (103.0)       (116.0)
Pre-COVID Mean                 0.434         0.429   0.420    0.445                 0.438         0.430       4,008.3 3,902.8         3,810.3      344.6       337.2         330.3
        Note: This table shows estimates of Equation 3b as described in the text for math scores, reading scores, enrollment counts, and absenteeism counts. Math scores
        and reading scores are measured in standard deviations of the 2019 test score distribution. Enrollment and absenteeism are presented in student-years, where
        enrollment is measured in the fall semester and absenteeism measures whether or not a student misses more than 10% of school days within a school year. We
        present estimates for three bandwidths, 45, 50, and 55 formula children from the 5% poverty threshold using a first-degree polynomial specification. We present a
        linear combination of all post-COVID coefficients as well as the SY 2017 to SY 2019 pre-COVID mean for all districts of the dependent variable at the bottom of
        the table. Regressions are weighted by SY 2020 enrollments and standard errors clustered by district are shown in parentheses.

        * p<0.05;** p<0.01; *** p<0.001




                                                                                       65
             Appendix Table 10: Robustness of Estimates of the Impact of Qualifying for Additional ESSER Funds on Measures of School District
             Performance to Second-Degree Polynomials
                                     Math                               Reading                        Enrollment                   Absenteeism
                             (1)      (2)        (3)         (4)          (5)         (6)         (7)      (8)       (9)       (10)      (11)     (12)
Qualification ∙ I{2023}    -0.0424 -0.0591 -0.0532 -0.0541           -0.0691*       -0.0545      101.1    129.0     152.7
                          (0.0324) (0.0312) (0.0298) (0.0338)         (0.0317)     (0.0301)     (104.2) (97.61) (92.56)
Qualification ∙ I{2022}    -0.0247 -0.0259 -0.0249 -0.0547* -0.0667** -0.0617**                  87.55    125.4    147.3*     -165.1 -336.7* -349.1*
                          (0.0293) (0.0282) (0.0275) (0.0255)         (0.0244)     (0.0232)     (84.38) (77.13) (72.16) (100.3) (139.1) (165.6)
Qualification ∙ I{2021}                                                                          58.65   114.7* 144.1**       79.76     186.9    186.4
                                                                                                (51.84) (53.10) (54.98) (145.0) (147.6) (143.7)
Qualification ∙ I{2020}                                                                                                       96.92     102.1    84.96
                                                                                                                             (96.07) (89.62) (80.06)
Second-Degree
Polynomial                    Y        Y          Y           Y            Y           Y           Y        Y         Y         Y         Y        Y
LEA FEs                       Y        Y          Y           Y            Y           Y           Y        Y         Y         Y         Y        Y
Year FEs                      Y        Y          Y           Y            Y           Y           Y        Y         Y         Y         Y        Y
Observations                5,991   6,547      7,036       5,427        5,962        6,459      30,751   32,732    34,762    24,311    25,910   27,521
R2                          0.914   0.915      0.916       0.908        0.908        0.912       0.997    0.997     0.997     0.691     0.689    0.689
Band                          45      50         55          45           50           55          45       50        55        45        50       55
Combined Coefficient       -0.0672 -0.0850 -0.0781 -0.1088             -0.1358      -0.1162     247.34   369.08    444.11     11.54    -47.67   -77.77
                          (0.0590) (0.0565) (0.0546) (0.0563)         (0.0531)     (0.0505)    (220.89) (203.95) (191.51) (226.16) (202.79) (172.91)
Pre-COVID Mean              0.434   0.429      0.420       0.445        0.438        0.430      4008.3   3902.8    3810.3     344.6     337.2    330.3
            Note: This table shows estimates of Equation 3b as described in the text for math scores, reading scores, enrollment counts, and absenteeism counts. We present
            estimates for three bandwidths, 45, 50, and 55 formula children from the 5% poverty threshold using a second-degree polynomial specification. We present a linear
            combination of all post-COVID coefficients as well as the SY 2017 to SY 2019 pre-COVID mean for all districts of the dependent variable at the bottom of the
            table. Regressions are weighted by SY 2020 enrollments and standard errors clustered by district are shown in parentheses.

            * p<0.05;** p<0.01; *** p<0.001




                                                                                           66
Appendix Table 11: Estimates of the Impact of Qualifying for Additional ESSER Funds on
Counts and Shares of Test Takers by Socioeconomic Group
                                                    Counts              Shares
                                                       Soc.     Not       Not
                                             All      Adv.      S.A.      S.A.
                                             (1)        (2)      (3)       (4)
              Qualification ∙ I{2023}      -51.11     118.8   -169.9 -0.006
                                          (430.4) (284.6) (188.6) (0.016)
              Qualification ∙ I{2022}      70.45     -159.5 229.9*       0.000
                                          (177.6) (220.3) (99.54) (0.015)
              Qualification ∙ I{2021}     1292.9      832.4    460.5     0.040
                                         (1103.7) (836.4) (356.0) (0.034)
              First Degree Polynomial         Y          Y        Y         Y
              LEA FEs                         Y          Y        Y         Y
              Year Fes                        Y          Y        Y         Y
              Observations                10,152    10,152 10,152 10,152
              R2                           0.978      0.959    0.863     0.726
              Band                            50        50       50        50
              Combined Coefficient         1,312       792      521       0.03
                                            (748)     (663)    (378)     (0.06)
              Pre-COVID Mean              4,053.0 3,159.8 893.2          0.279
Note: This table shows estimates of Equation 3b as described in the text for counts of total test takers, counts of
socioeconomically advantaged test takers, counts of not-socioeconomically advantaged test takers, and the share of
non-socioeconomically advantaged test takers out of all test takers. We present estimates for 50 formula children from
the 5% poverty threshold. We present a linear combination of all post-COVID coefficients as well as the SY 2017 to
SY 2019 pre-COVID mean for all districts of the dependent variable at the bottom of the table. Regressions are
weighted by SY 2020 enrollments and standard errors clustered by district are shown in parentheses.

* p<0.05; ** p<0.01; *** p<0.001




                                                         67
Appendix Table 12: Estimates of the Impact of Qualifying for Additional ESSER Funds on
Percent Changes in Private School District Enrollment Around Public Schools Between SY 2022
and SY 2020
                                             Percent Change in Enrollment from
                                                            SY 2020
                                              5 Mile Radius      5-10 Mile Radius
                                                         In                  In
                                              All     Grade        All     Grade
                                              (1)       (2)        (3)      (4)
             Qualification Dummy            -0.119    -0.256     0.054     0.109
                                           (0.138) (0.266) (0.076) (0.109)
             First Degree Polynomial           Y         Y          Y        Y
             Observations                   1,597     1,543      2,159     2,129
             R2                             0.008     0.004      0.000     0.001
             Band                             50        50         50        50
             Pre-COVID Enrollment          1,669.2 1,371.8 4,668.3 3,875.5
Note: This table shows estimates of Equation 4 as described in the text for percent changes in private school enrollment
between SY 2022 and SY 2020 (measured in October of the fall semester). These data come from the Private School
Universe Survey (Institute of Education Sciences, 2024b). Private school enrollment is calculated based on proximity
to the public school district of interest. We sum all private school enrollments and private school enrollments in
overlapping grades to the reference public school in a 5 mile radius and then a 5 mile to 10 mile radius donut in SY
2020 and SY 2022. Then, we construct the percentage change across the two school years. Column 1 shows estimates
for percent changes in all private school students within a 5 mile radius of the reference public school. Column 2
shows estimates for percent changes in private school enrollments in grades that overlap with the reference public
school district within 5 miles of the reference public school district. Column 3 shows estimates for percent changes in
all private school students within a 5 mile to 10 mile radius donut from the reference public school. Column 4 shows
estimates for percent changes in private school enrollments in grades that overlap with the reference public school
district within a 5 mile to 10 mile radius donut from the reference public school district. We present estimates for 50
formula children from the 5% poverty threshold. We present a linear combination of all post-COVID coefficients as
well as the SY 2017 to SY 2019 pre-COVID mean for all districts of the dependent variable at the bottom of the table.
Regressions are weighted by SY 2020 enrollments and standard errors clustered by district are shown in parentheses.

* p<0.05; ** p<0.01; *** p<0.001




                                                          68
Appendix Table 13: Estimates of the Impact of Qualifying for Additional ESSER Funds on Area
Average Daily COVID Cases per 100,000 Residents in SY 2021
                                      Dep. Var: Area Daily Cases per 100k in 2021
                                  (1)       (2)       (3)      (4)        (5)     (6)
     Qualification Dummy        -0.896    0.274     -1.584    0.772     -1.539   0.190
                               (0.888) (1.317) (0.834) (1.260) (0.804) (1.229)
     First-Degree
     Polynomial                    Y                   Y                   Y
     Second-Degree
     Polynomial                              Y                  Y                  Y
     Observations                4,412    4,412     4,695     4,695     4,983    4,983
     R2                         0.0224 0.0238 0.0169 0.0215 0.0180 0.0206
     Band                          45       45        50        50        55       55
     Control Mean                24.40    24.40     24.25     24.25     24.28    24.28
Note: This table shows estimates of Equation 4 as described in the text for area average daily case rates of COVID-
19. We show first- and second-degree polynomial estimates using bandwidths of 45, 50, and 55 formula children for
SY 2021. Below-cutoff means of the dependent variable are presented at the bottom of the table. We show robust
standard errors in parentheses. We weight estimates by SY 2020 enrollments.

* p<0.05;** p<0.01; *** p<0.001




                                                        69
       Appendix Table 14: Estimates of the Impacts of Qualifying for Additional ESSER Funds on
       School District Debts and Assets

                                                                                                             Asset Types
                                                                       Debt.          Tot.        Sinking        Bond              Other
                       Tot. Rev.      Tot. Exp.       Tot. Debt       Interest      Assets          Fund         Fund             Assets
                           (1)           (2)             (3)            (4)            (5)           (6)          (7)                (8)
Qual. ∙ I{2022}          -357.9         192.3           969.0          78.12         -717.6        -227.0       -271.2             -219.4
                        (214.3)        (555.6)         (911.9)        (98.55)       (778.7)       (148.6)      (676.9)            (313.0)
Qual. ∙ I{2021}          -182.1         454.0           129.8          23.53         -816.1        -156.8       -718.2             58.79
                        (245.5)        (463.4)         (764.3)        (40.62)       (716.1)       (166.9)      (638.8)            (279.3)
Qual. ∙ I{2020}          -3.504         99.22           710.6          102.9         -327.7        -182.7       -186.4             41.40
                        (173.5)        (369.3)         (628.4)        (80.95)       (597.9)       (155.3)      (584.8)            (262.3)
First-Degree
Polynomial                  Y         Y          Y        Y         Y         Y         Y         Y
LEA FEs                     Y         Y          Y        Y         Y         Y         Y         Y
Year FEs                    Y         Y          Y        Y         Y         Y         Y         Y
Observations             28,056    28,056     28,056   28,056    28,056    28,056    28,056    28,056
R2                        0.898     0.829      0.846    0.749     0.714     0.652     0.443     0.866
Band                        50       50          50       50       50         50       50         50
Combined                 -543.51   745.45    1,809.39 204.54 -1,861.41 -566.45 -1,175.74 -119.21
Coefficient             (486.14) (1,217.20) -2,064.37 (185.12) (1,835.18) (419.29) (1,688.49) (601.33)
Pre-COVID Mean         18,015.10 17,992.80 11,488.10 457.60     7,497.60   561.00   1,740.90 5,195.70
       Note: This table shows estimates of Equation 3b as described in the text for total revenues per pupil, total expenditure
       per pupil, total outstanding debt per pupil, interest payments on debt per pupil, total assets per pupil, sinking fund
       assets per pupil, bond fund assets per pupil, and other assets per pupil. Total debt is the face value (par value) of
       outstanding debts of all maturities at the end of the fiscal year. Total assets are made up of sinking funds, unspent
       bond funds, and other assets, which include general funds and other investments made by the district. Per pupil
       measures use SY 2020 enrollments. We present estimates for 50 formula children from the 5% poverty threshold and
       a first-degree polynomial specification. We present a linear combination of all post-COVID coefficients as well as the
       SY 2017 to SY 2019 pre-COVID mean for all districts of the dependent variable at the bottom of the table. Regressions
       are weighted by SY 2020 enrollments and standard errors clustered by district are shown in parentheses.

       * p<0.05;** p<0.01; *** p<0.001




                                                                 70
Appendix Table 15: Estimates of the Impacts of Qualifying for Additional ESSER Funds on
Total Compensation Per Employee by Type

                                                     Tot. Staff Teachers Non-Teachers
                                                         (1)       (2)        (3)
                   Qualification ∙ I{2022}            -193.70   -173.80   -1,363.10
                                                       (1,232)   (1,260)    (3,990)
                   Qualification ∙ I{2021}            3,989.40 3,178.80   7,979.20*
                                                       (3,003)   (3,876)    (4,006)
                   Qualification ∙ I{2020}           2,097.90* 1,635.90    6,065.30
                                                        (948)     (917)     (3,280)
                   First-Degree Polynomial                Y         Y          Y
                   LEA FEs                                Y         Y          Y
                   Year FEs                               Y         Y          Y
                   Observations                        27,978    27,854     27,749
                   R2                                   0.876     0.873      0.715
                   Band                                   50        50         50
                   Combined Coefficient                 5,894     4,641     12,682
                                                       (3,971)   (4,836)    (9,364)
                   Pre-COVID Mean                      84,813   100,243     75,133
Note: This table shows estimates of Equation 3b as described in the text for total compensation per employee for three
employee categories. Total compensation includes both benefits and salaries. Column 1 presents estimates for all staff,
then Column 2 presents estimates for teachers, and finally Column 3 presents estimates for non-teachers. All columns
use both compensation and staff counts that update in each year. These staff counts are the same full time equivalent
counts used above on a per pupil basis. We present estimates for 50 formula children from the 5% poverty threshold
and a first-degree polynomial specification. We present a linear combination of all post-COVID coefficients as well
as the SY 2017 to SY 2019 pre-COVID mean for all districts of the dependent variable at the bottom of the table.
Regressions are weighted by SY 2020 enrollments and standard errors clustered by district are shown in parentheses.

* p<0.05;** p<0.01; *** p<0.001




                                                          71


File and source

File

intergovernmental-grants-to-school-districts-and-educational-outcomes-during-the.pdf

Size

1,559,853 bytes

SHA-256

a139b0159140d48e6ca577f13e7890adc6d2d76a6b68714964f8f44557d3f416

Our copy

intergovernmental-grants-to-school-districts-and-educational-outcomes-during-the.pdf

Original

www.nber.org