Collinson et al., The Effects of Emergency Rental Assistance During the Pandemic: Evidence from Four Cities (NBER w32463 (rev. Oct 2025), 2024-05)
Summary
NBER Working Paper 32463, dated May 2024 and revised October 2025, by Robert Collinson, Anthony A. DeFusco, John Eric Humphries, Benjamin J. Keys, David C. Phillips, Vincent Reina, Patrick S. Turner and Winnie van Dijk. The paper evaluates five emergency rental assistance programs distributing over $200 million through lottery. The abstract says assistance led to increases in rent payment and reduced concerns about eviction, with suggestive improvements in self-reported mental and physical health. It finds little effect on housing stability or financial distress and lists eviction moratoria, expanded safety net programs and market softening favoring tenants as explanations.
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
Cite this document
“Collinson et al., The Effects of Emergency Rental Assistance During the Pandemic: Evidence from Four Cities (NBER w32463 (rev. Oct 2025), 2024-05).”
Original at www.nber.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/collinson-et-al-the-effects-of-emergency-rental-assistance-during-the-60bf343960/
, SHA-256 f7cd88a1a21dae18105ddaa29d7d7447582611016faab9d77787cc362816fc18.Full text
NBER WORKING PAPER SERIES
THE EFFECTS OF EMERGENCY RENTAL ASSISTANCE DURING THE PANDEMIC:
EVIDENCE FROM FOUR CITIES
Robert Collinson
Anthony A. DeFusco
John Eric Humphries
Benjamin J. Keys
David C. Phillips
Vincent Reina
Patrick S. Turner
Winnie van Dijk
Working Paper 32463
http://www.nber.org/papers/w32463
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
May 2024, Revised October 2025
We thank Blanca Garcia, Emily Davis, Will Day, Kenny Hofmeister, Jacqueline Kelley-
Cogdell, Ayush Sinha, Stephen Stapleton, Peter Williams, and Seth Zissette for excellent
research assistance. DeFusco gratefully acknowledges financial support from the Peter
G. Peterson Foundation Pandemic Response Policy Research Fund. Collinson, Phillips,
and Turner acknowledge support from the Wilson-Sheehan Lab for Economic
Opportunities (LEO). J-PAL North America and LEO provided financial support for
surveys. Special thanks to King County employees Jeremey Breicher-Haimson, Christina
McHugh, Emily Reimal, and Jesse Warren for lending expertise on both data and policy;
the views expressed here are those of the authors and do not necessarily represent the
views of King County. Keys thanks the research sponsors of Wharton’s Zell/Lurie Real
Estate Center for support. Reina thanks Claudia Aiken and Sydney Goldstein for their
research assistance and the Stoneleigh Foundation for their support for surveys, and Ann
Sewell, Anna Ortega, Marcella H. DeShurley and others in the City of Los Angeles for
their expertise and data; the views expressed here are those of the authors and do not
necessarily represent the views of the City of Los Angeles. Humphries and Van Dijk
thank the University of Chicago Women’s Board, J-PAL North America, William H.
Block, Crown Family Philanthropies, the Polk Bros. Foundation, and the Yale Tobin
Center for Economic Policy for financial support; Daniel Hertz, Erendira Rendon, and
Paul Williams for sharing their expertise on program design and implementation;
Carmelo Barbaro, Janelle Blackwood, Emma Boczek, Shannon Bradford, Anne Driscoll,
Kelly Hallberg, Kirsten Jacobson, Emily Metz, Thomas Rietz, and Preethi Varma at the
University of Chicago Inclusive Economy Lab for feedback and research support. We
thank Ron Borzekowski, Ben Hyman, Rajashri Chakrabarti, Paul Goldsmith-Pinkham
(discussant), and seminar participants at the ASSA meetings, the Housing Initiative at
Penn, the Urban Economics Association meetings, and the NBER Household Finance
working group for helpful comments. This project was approved by IRB’s at the
University of Chicago (IRB20-1673), the University of Notre Dame (IRB20-12-6355),
and the University of Pennsylvania (843183). Analyses from two sites in our study were
registered with the American Economic Association’s RCT Registry and the study IDs
are AEARCTR-0007731 (Chicago) and AEARCTR-0007167 (King County). 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.
© 2024 by Robert Collinson, Anthony A. DeFusco, John Eric Humphries, Benjamin J.
Keys, David C. Phillips, Vincent Reina, Patrick S. Turner, and Winnie van Dijk. 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.
The Effects of Emergency Rental Assistance During the Pandemic: Evidence from Four Cities
Robert Collinson, Anthony A. DeFusco, John Eric Humphries, Benjamin J. Keys, David C.
Phillips, Vincent Reina, Patrick S. Turner, and Winnie van Dijk
NBER Working Paper No. 32463
May 2024, Revised October 2025
JEL No. H00, I38, J01, R38
ABSTRACT
Short-term rental assistance expanded to unprecedented scale during the COVID-19 pandemic.
We evaluate five programs distributing over $200 million through lottery, using administrative
and survey data to assess effects on rent payment, housing stability, financial distress, and
health. Assistance led to increases in rent payment and reduced concerns about eviction, with
suggestive improvements in self-reported mental and physical health. In contrast with pre-
pandemic emergency rental assistance, we find little effect on housing stability or financial
distress. Explanations for these muted effects include: eviction moratoria weakening the link
between rent and displacement, expanded safety net programs, and market softening favoring
tenants.
Robert Collinson David C. Phillips
University of Notre Dame University of Notre Dame
Department of Economics Department of Economics
and NBER dphill12@nd.edu
rcollin8@nd.edu
Vincent Reina
Anthony A. DeFusco University of Pennsylvania
University of Wisconsin - Madison vreina@upenn.edu
Finance Department
and NBER Patrick S. Turner
anthony.defusco@wisc.edu University of Notre Dame
Department of Economics
John Eric Humphries Patrick.turner@nd.edu
Yale University
Department of Economics Winnie van Dijk
and NBER Yale University
johneric.humphries@yale.edu Department of Economics
and NBER
Benjamin J. Keys winnie.vandijk@yale.edu
University of Pennsylvania
The Wharton School
and NBER
benkeys@wharton.upenn.edu
Full survey instruments are available at https://github.com/robcollinson/covid_era_surveys/
1 Introduction
The COVID-19 pandemic created new health risks and upended economic activity. The US
government responded by rolling out policies to provide relief to households through tools
used during other moments of crisis, including direct cash assistance (“Economic Impact
Payments”) and the expansion of traditional transfer programs (such as Unemployment In-
surance benefits). However, unlike previous economic crises, policymakers also allocated
an unprecedented amount of resources to rental assistance. Federal funding for Emergency
Rental Assistance (ERA) totaled over $50 billion, representing an amount roughly equiva-
lent to the annual budget for all federal rental assistance programs prior to the pandemic
(Collinson et al., 2019).
Despite the unprecedented expansion of funding for ERA, there is little, if any, evidence
available on its effectiveness as a relief mechanism during the pandemic. This lack of ev-
idence is due to several factors. First, there was no centralized data collection on ERA
program administration. Second, the ability to track outcomes such as rent payment, evic-
tion, homelessness, and households’ health and financial situations is limited by the absence
of comprehensive national administrative data. Third, comparing assisted and unassisted
individuals is likely affected by selection bias, calling into question any causal interpretation.
In normal economic times, theory and prior evidence suggest ERA may affect tenant
outcomes through several channels. Payment of rent arrears should prevent eviction court
proceedings, maintaining housing stability and avoiding the documented negative conse-
quences of forced moves on credit, employment, and health (Collinson et al., 2024). By
reducing financial strain, assistance may also improve mental health and allow households
to meet other financial obligations. Indeed, pre-pandemic ERA programs providing similar
assistance amounts have been shown to effectively prevent homelessness (Evans et al., 2016;
Phillips and Sullivan, 2025).
However, the pandemic created unique policy and market conditions that could fun-
damentally alter the expected effects of rental assistance. Unprecedented expansions of
unemployment insurance, direct cash payments, and other safety net programs may have
diminished the marginal impact of additional rental assistance. Simultaneously, federal and
local moratoria were placed on eviction for nonpayment of rent. These moratoria did not
necessarily eliminate housing instability, as landlords could still pursue evictions for lease
violations, decline lease renewals, or apply informal pressure. Nevertheless, they likely weak-
ened the typical link between rental arrears and displacement, a primary mechanism through
which ERA preserves housing stability. Finally, the pandemic also shifted rental market
dynamics, with falling rents and rising vacancy rates potentially making landlords more
1
accommodating, regardless of formal assistance programs.
In this paper, we provide the first empirical evidence on the causal impacts of ERA
programs deployed in response to the pandemic, allowing us to test how these unique cir-
cumstances affected program effectiveness. We study five local ERA programs, each over-
subscribed and selecting participants through a lottery, providing $1,000 to $3,400 to house-
holds in four major urban areas: Chicago, Harris County (Houston), King County (Seattle),
and Los Angeles. Collectively, these programs distributed approximately $200 million and
received more than 200,000 applications for assistance. Figure I illustrates the backdrop
against which they were implemented. Their timing coincided with the initial economic im-
pact of the pandemic, the peak of reported rates of rental arrears, and the roll-out of other
timely government relief initiatives. Because these local programs provided assistance well
before a larger federal ERA funding round, they are well-suited for evaluating ERA as a
crisis-response mechanism supplementing more traditional relief programs.
To learn about the causal impact of receiving assistance, we use exogenous variation
from oversubscribed assistance lotteries in the five programs.1 Using data on applications to
these programs linked to survey and administrative data, we estimate their causal impact on
policy-relevant outcomes related to housing stability, financial distress, and mental health.
We use credit bureau records and consumer reference data to track the evolution of financial
stability and housing outcomes. We augment these data with novel harmonized surveys to
capture results that were not easily measured in administrative sources. For select sites, we
also incorporate eviction court filings and homelessness system records to examine extreme
housing instability.
Receipt of assistance increased rent payments and reduced self-reported anxiety. Using
harmonized surveys fielded across four of the lotteries, we find that those who received
assistance were 6–13 percentage points (8–36 percent) more likely to pay their rent in full
in the months shortly after the lottery. The estimated effects on rent payment are similar
across sites but are somewhat larger for communities that offered more generous assistance
and made payments directly to landlords.
We use the same surveys to examine the impacts of assistance on measures of mental
health and on concerns about eviction. Receipt of assistance reduced self-reported anxiety
by an average of 3.2 percentage points (6 percent) across sites. We additionally find that
assistance reduced concern about eviction, with a 4.7 percentage point reduction (16 percent)
across sites.
Despite the policy objective of keeping individuals and families in their homes, we find
1
Our analysis follows a pre-specified plan. See AEA RCT Registry #0007731 and #0007167 and Appendix
C for details.
2
no consistent effects of the assistance on the likelihood that applicants moved. We analyze
applications to all five programs linked to two separate administrative sources to draw this
conclusion. Using both a panel of linked credit data and address histories from consumer
reference data, we find little evidence to suggest that assistance made individuals less likely
to move from their application address in either the two months or the year after applying
for assistance. Overall, the move rates in our samples appear similar for the period leading
up to the lottery and the post-lottery period.
Next, using linked credit reports, we examine whether the assistance affected short-
and longer-run measures of financial distress. While the assistance had small short-run
impacts on some credit measures, these effects are not consistently present across sites and
did not persist over time. Averaging across sites, individuals assisted through the lottery
had approximately $74 less in collections in the two months after applying for assistance.
However, these effects appear to fade over time. The financial characteristics of applicants
randomized into receiving assistance are largely indistinguishable ten months after the lottery
from those of applicants who missed out.
Our last set of estimates examines whether receiving assistance decreased applicants’
chances of facing severe housing instability, including eviction or homelessness. To measure
the impacts on homelessness, we link the application data for three of our lotteries to home-
lessness system data from King County (Washington) and Chicago (Illinois). We do not find
consistent evidence that receiving assistance reduced the likelihood of applicants appearing
in the homelessness system.2 We then investigate whether assistance reduced the probability
of a new eviction filing in Chicago and Harris County, both shortly after assistance and after
the national Centers for Disease Control (CDC) eviction moratorium was lifted. We find
that the assistance had little effect on eviction filings and that baseline rates of new filings
remained relatively low for applicants not assisted through the lottery.
While we estimate generally muted effects of ERA, we nevertheless find that it was well
targeted to financially distressed households. The recipients of assistance across our four
sites appear more financially distressed than the typical renter in each community. We find
that compared to the general population or other renters in the same community, those who
received assistance had 33–141 points lower credit scores, higher collections balances, and
higher levels of debt. Assisted households were also from higher-poverty neighborhoods and
were more likely to be minorities. We reach similar conclusions when comparing those who
received ERA assistance to those who received Economic Impact Payments (EIP).
Our results suggest that receipt of assistance was well targeted, helped households pay
2
The only group for whom we find such evidence are applicants to a program in Chicago targeting
individuals who might have limited access to other parts of the pandemic safety net.
3
their rent in the short run, and led to modest improvements in concerns about eviction,
with suggestive evidence that it improved mental and physical health. However, assistance
did not substantially change applicants’ housing or financial stability. Furthermore, when we
investigate treatment effect heterogeneity across a wide range of individual characteristics, as
well as across sites, we find little evidence to support heterogeneous effects of the assistance.
These conclusions contrast with prior research on pre-pandemic ERA programs that provided
similarly-sized assistance to renters facing household-level shocks. Evans et al. (2016) and
Phillips and Sullivan (2025) find large effects on recipients’ housing stability. This contrast
is particularly striking given that the programs share similar design features and target
populations.
What forces mitigated the effectiveness of pandemic ERA? We explore three potential
channels: eviction moratoria, the expansion of other safety net programs, and equilibrium
responses in the rental market. Eviction moratoria likely reduced the marginal value of
assistance in preventing displacement. However, we find that concern about future eviction
was elevated during the period when the ERA programs we study distributed their funds (see
Figure I). Moreover, the fact that we find little heterogeneity across sites, despite differences
in either the interpretation of the federal moratorium or the existence of extended local
moratoriums, suggests that this cannot be the only explanation.
Next, we examine the role of an expansion in other forms of assistance. Our linked credit
data reveal that individuals who applied for assistance experienced rising credit scores and
falling debt in the lead-up to application (in line with Fulford, 2023). These patterns likely
reflect a mix of broad-based improvements in macroeconomic conditions after the initial
COVID shock and the effects of other expanded social insurance and safety-net programs
during the pandemic.3 The improvement in household balance sheets continues post-lottery
but is similar for both treatment and control applicants, consistent with our findings of small
causal impacts on the credit measures. Two additional pieces of evidence point to the impor-
tance of other safety-net expansions to understanding our small estimated impacts. First,
lottery applicants with lower pre-existing attachments to the safety-net appear to experience
larger reductions in housing instability from assistance than those already connected to the
safety-net. Second, applicants to the Chicago TRP lottery, which sought to serve individuals
with less access to other safety-net help (such as undocumented or mixed-status families),
experience a statistically significant reduction in contact with the homelessness system.
Lastly, we consider equilibrium responses in the rental market. The early phase of the
3
See, for example Blanchet et al. (2022), who observe that “After accounting for taxes and cash and
quasi-cash transfers, disposable income for adults in the bottom 50% was 20% higher in 2021 than in 2019.”
See also Han et al. (2020) and Raphael and Schneider (2023).
4
pandemic saw sharp decreases in rent prices and increases in vacancy rates. For landlords,
evicting a tenant is likely a less attractive option during a widespread downturn than during
normal times, when vacancies can be more easily filled at higher rents. Therefore, renegoti-
ation or temporary agreements between landlords and tenants may be sufficient to forestall
eviction without additional emergency assistance. Consistent with this mechanism, our sur-
veys show very high rates of reported landlord agreements (over 70 percent) among tenants
who are having trouble paying their rent. This evidence is in line with other pandemic-era
surveys of landlords that find substantial increases in reported rental forbearance in 2020,
relative to pre-pandemic levels (Decker, 2021; de la Campa and Reina, 2023).
Our paper contributes to a large literature studying the effects of various policies seeking
to stabilize housing markets during economic downturns. Much of this literature has focused
on the owner-occupied segment of the market, where frictions in mortgage financing created
substantial turmoil during the global financial crisis (Agarwal et al., 2017; Piskorski and Seru,
2018; Ganong and Noel, 2020; Agarwal et al., 2022). In contrast, our paper is one of the
first to study the effects of large-scale relief programs targeted specifically at renters during
a time of macroeconomic distress. This emphasis is particularly relevant because renters
reported significantly higher rates of missing housing payments than owner-occupants (15
percent compared to 9.5 percent) and higher risk of losing their home (7 percent compared
to 2 percent) during the pandemic (U.S. Census Bureau, 2020c).
Other work has studied COVID-19 ERA, mostly focusing on the federal ERA program
that was instituted near the tail end of the pandemic. Some of these studies have connected
COVID-19 ERA to improvements in housing outcomes, financial well-being, and mental
health (Airgood-Obrycki, 2022; Reina and Lee, 2023). However, these studies rely on de-
scriptive comparisons of households that received assistance to those that did not, whereas
we are able to leverage randomized receipt of assistance generated via lottery. Rafkin and
Soltas (2024) find similarly small ERA impacts on eviction using event-study methods.
This paper also connects to the literature on housing instability and rental assistance.
Prior work examines the effects of evictions (Collinson et al., 2024), landlord foreclosure (Di-
amond et al., 2019), and short-term housing assistance for recipients experiencing household-
level shocks (Evans et al., 2016; Phillips and Sullivan, 2025). We provide the first rigorous
evidence on the causal effects of ERA during a time of aggregate distress and concomitant
large-scale policy responses. We also study a wider range of outcomes including housing
stability, financial strain, and the mental and physical health of program applicants.
We contribute to a growing literature analyzing the impacts of economic policies during
the COVID-19 pandemic by providing the first comprehensive analysis of the effects of the
emergency assistance distributed to renters. Chetty et al. (2023) evaluate a range of other
5
pandemic economic policies using nearly real-time data on spending from private sector
sources. Ganong et al. (2022) examine the impacts of pandemic unemployment insurance
on spending and job-finding, while Karger and Rajan (2021) study the effect of Economic
Impact Payments on spending.4 While past work has sought to estimate the impact of the
eviction moratoria during the pandemic on COVID-19 infections and mortality (Jowers et
al., 2021) and on credit and mental health outcomes (An et al., 2021), we provide the first
large-scale evidence on the causal impact of a key part of the immediate economic response
across four large urban areas. In contrast to the existing work, which uses county-level
data, we also use our individual- and household-level data to study how renters’ financial
circumstances developed throughout the pandemic.
Finally, our work relates to recent studies of pandemic cash assistance. Jaroszewicz et al.
(2022) and Pilkauskas et al. (2022) examine one-time cash payments ($500-$2,000 and $1,000,
respectively) and find minimal effects on spending, material hardship, and mental health. We
study rental assistance—a major component of pandemic relief—across four metropolitan
areas with larger samples. While we find similar limited effects on bill payment and financial
strain, we do find some evidence of mental health improvements, in contrast to these studies.
2 Background
2.1 Program Details
We study five emergency rental assistance lotteries administered in four locations during
2020. The programs launched at different points during the pandemic’s first year: Chicago’s
Department of Housing (DOH) program operated earliest (applications March 27-April 5,
2020), followed by Los Angeles (July 13-17), The Resurrection Project (TRP) in Chicago
(applications opened June 22), King County (August 20-December 4), and Harris County
(applications closed November 5).
Most programs provided similar amounts of assistance: Chicago DOH and TRP each
distributed $1,000 grants paid directly to tenants, Harris County provided $1,200 in direct
cash assistance, and Los Angeles offered $2,000 per household distributed either directly to
the tenant or their landlord. King County was notably more generous, providing an average
of three months’ back rent paid to landlords (with a maximum of six months), though
landlords had to agree to substantial conditions including forgiving rental debt beyond six
months and not raising rent through March 2022.
4
Recent studies have also analyzed the short-term effects of the 2021 expansion of the Child Tax Credit
(CTC) on household financial and mental well-being (Collyer et al., 2022; Glasner et al., 2022; Kovski et al.,
2023; Pignatti and Parolin, 2023; Pilkauskas et al., 2023).
6
Income eligibility requirements varied across sites. Chicago DOH and Harris County lim-
ited eligibility to households earning no more than 60 percent of area median income (AMI),
King County set a stricter threshold of 50 percent AMI, and Los Angeles allowed incomes up
to 80 percent AMI. Chicago TRP was distinctive in explicitly targeting populations ineligi-
ble for CARES Act assistance: while nominally open to all Chicago residents with incomes
below 300 percent of the federal poverty line, applications from undocumented individuals,
mixed-status families, dependent adults, and returning residents were particularly encour-
aged. This focus on reaching undocumented populations represents a key contrast with the
other programs we study.
Complete program details, including application procedures and lottery mechanisms, are
provided in Appendix A.
2.2 Policy Context and Program Timing
Figure I illustrates the timing of the five programs we study relative to the broader macroe-
conomic and policy environment of the pandemic. Panel A shows that unemployment spiked
dramatically in early 2020, reaching nearly 15 percent before gradually declining. Panel B
documents the impact of federal relief initiatives on disposable household income for the bot-
tom half of the income distribution, comparing income before and after taxes and transfers.
While factor income declined sharply during the pandemic, post-tax disposable income—
which includes Economic Impact Payments, Child Tax Credit, Earned Income Tax Credit,
and expanded Unemployment Insurance benefits—actually increased substantially for lower-
income households. These federal transfers were implemented around the same time as the
programs we study.
The federal government also provided funding specifically for emergency rental assistance,
but in a later phase of the pandemic. As Figure I makes clear, all five programs we study
distributed assistance well before the large federal ERA1 and ERA2 funding rounds began
in mid-2021. Given these differences in timing, as well as differences in implementation and
assistance amounts, our results are not necessarily representative of the impact of the federal
ERA programs. However, the lotteries we study provide arguably the best available setting
for evaluating ERA as a crisis-response mechanism during the most immediate and highest-
intensity economic impact of the pandemic. Panel C shows the scale of renter distress during
this period: approximately 10 to 13 million households—representing 23-30 percent of all
renter households—reported being behind on rent when these programs were rolled out,
according to Census Household Pulse survey data. This number fell sharply to just above
7 million households (approximately 16 percent) by spring 2021 and never again reached
7
the peak levels seen during the programs’ implementation period.5 Panel D documents how
eviction filings varied across our study sites during this time.
National and local eviction moratoria also shaped the environment in which these pro-
grams operated, though their practical impact varied considerably. The CARES Act insti-
tuted a 120-day national moratorium on eviction filings for properties with federal financing
(March 27-July 24, 2020), followed by a broader CDC moratorium (September 4, 2020-
July 31, 2021, with a short second moratorium August 3-26, 2021, that the Supreme Court
struck down). These national moratoria covered nonpayment evictions but allowed evictions
for other causes. Their implementation also differed across locations due to inconsistent
judicial interpretation. Harris County proved particularly notable for weak enforcement.
Despite the CDC moratorium, the county averaged over 500 weekly eviction filings during
this period, approximately 60 percent of pre-pandemic volume, amid widespread concerns
that local judges were not honoring federal protections.
All four study sites also enacted local moratoria, but their strength and duration var-
ied: Los Angeles maintained its moratorium through June 2022, Chicago and the state of
Washington extended protections through October 2021 (with Seattle and other Washington
cities extending to January 2022), while Harris County’s brief state moratorium (March-May
2020) ended early.
3 Data
We link administrative records from each of the five programs to five types of data sources.
This section provides details on each of these data sources.
3.1 Credit Bureau Data
Our analysis of financial distress uses individual-level credit reporting data from Experian,
one of the three major credit bureaus. The data contain a total of twelve monthly snapshots
of consumer credit profiles observed bimonthly beginning at the end of January 2020 and
running until the end of November 2021. These data allow us to track the evolution of credit
outcomes throughout the early pandemic. They include information on credit scores, debt
5
Figure I includes three related measures of renter households’ distress in Panel C: the number of house-
holds behind on rent, those unconfident in their ability to pay next month’s rent, and those in rental arrears
reporting themselves at risk of eviction. The data show lower levels but very similar trends for the latter two
measures. All three measures strongly suggest that tenants had the most trouble paying rent in the early
phase of the pandemic.
8
levels and delinquency status for all major forms of consumer debt.6 In addition to these
credit variables, the data also include the ZIP code of residence at the time of each snapshot,
which we use to infer applicants’ residential mobility.
Experian matched the program application data to credit reporting data for all sites
using the applicant’s name and address. All lottery participants were eligible to be matched,
except those in King County, where only survey respondents were eligible to be matched. To
protect applicant confidentiality, all personally identifiable information was removed from
the matched data before the latter were returned to the research team by Experian. Among
people eligible to be matched, match rates across all sites are generally high, ranging from
approximately 74 percent for Harris County to 81 percent for King County. Conditional
on being observed in the credit data, nearly all applicants are present in every monthly
snapshot. Our analysis therefore uses a balanced panel of credit outcomes.
3.2 Consumer Reference Data
As an alternative way to measure address changes, we use consumer reference data from
Infutor Data Solutions. Infutor combines various information from many sources (e.g., phone
bills, voter files, magazine subscriptions, and property deeds) to create a residential history
for most adults in the United States. The result is a database of exact addresses that crosses
state boundaries and includes start and end dates for each address. While these data were
originally created for commercial applications such as identity verification and marketing,
academics have recently used them to measure housing stability and migration in the wake
of rent control, eviction, and new housing construction (Diamond et al., 2019; Asquith et
al., 2023; Collinson et al., 2024). The Infutor data are well-suited for our study because they
provide a consistent measure of housing stability across several metropolitan areas. Because
the data are compiled from consumer records, they match with other data at a somewhat
lower rate than records from other administrative sources and tend to miss groups of people
more inclined to have shorter paper trails (e.g., young or Hispanic individuals). Even so, prior
work shows that the data can measure residential mobility in similar high-risk populations;
for example, address changes measured in Infutor data spike around the time people at risk
of homelessness request and are denied emergency financial assistance from a large call center
(Phillips, 2020).
6
Under the CARES Act, creditors were required to make a variety of hardship “accommodations” for
households directly impacted by COVID-19, including deferring or forbearing payments, or reporting ac-
counts as current, but required households to actively request relief (Fiano, 2020). Aggregate statistics
suggest that delinquency on household credit continued to be reported to the credit bureaus, except for
federal student loans, which were reported as current while payments were paused (Household Debit and
Credit Report, 2023).
9
We measure address changes among the people we are able to match to the Infutor data.
The Infutor records include name, date of birth, and address, and we match on combinations
of these identifiers in each location, depending on their availability in the application records
for a given site. We limit our analysis to individuals who have an address in Infutor with a
start date prior to their application for financial assistance. This process yields match rates
of 35 percent for Chicago DOH, 21 percent for Chicago TRP, 35 percent for Harris County,
46 percent for King County, and 50 percent for Los Angeles. For this sample, we identify
the address at which the person resided when applying for assistance and measure whether
Infutor later observes this person at a different address.
3.3 Survey Data
In addition to administrative data, our analysis uses surveys administered in the months
following treatment from four of the five programs studied. The surveys carried out in
Chicago, Harris County, and King County relied on largely harmonized survey instruments,
with questions that were similar but not identical across sites.7 When processing the survey
data, we construct outcomes to align as closely as possible across sites. However, the survey
questions used in Los Angeles were substantially different from the other sites. For this
reason, we report the results for Los Angeles separately in the appendix. Appendix Section
B provides a detailed discussion of how our survey-based outcomes are constructed.8
The amount of time that elapsed between assistance and the administration of the survey
also varied across sites. Appendix Table E.2 reports the survey dates for each site. The
surveys were administered one to six months after assistance was received. The Chicago
DOH survey was administered the month after treatment, while other sites, such as Harris
County, administered their surveys four to five months after assistance was received. For
some of the survey outcome variables, retrospective questions were used to improve alignment
across sites so as to account for differences in the elapsed time between the lottery and the
completion of the survey.
The bottom panel of Table I reports survey response rates at each site. Survey response
rates vary between 13 and 40 percent. For most sites, individuals who were offered assistance
are more likely to respond to surveys than the those who were not. Appendix Table E.5
7
This research began as three independent projects. Site-specific surveys were initially developed in-
dependently and in collaboration with partner organizations. The research teams eventually shared the
survey instruments and, to the extent possible, aligned them. The full survey instruments can be found at
https://github.com/robcollinson/covid_era_surveys/.
8
Even among the sites where survey questions were largely aligned, one notable exception is the “COVID
Positive” outcome, the Chicago DOH survey codes this outcome as one if the respondent selects “I got sick”
to the multiresponse question “How has the COVID-19 epidemic impacted you?” The other sites’ surveys
specifically ask if the respondent suspected that she was infected with COVID-19 or had a positive self-test.
10
shows that survey nonresponse leads to modest imbalances in baseline characteristics. As
we describe in the next section, our main analysis corrects for this imbalance by weighting
responses by the applicant’s inverse probability of responding to the survey as predicted from
baseline characteristics observed for all applicants (Wooldridge, 2007).
3.4 Evictions Data
We supplement the data sources above with eviction data from Harris County and Cook
County. These data, and the homelessness data described in the next subsection, allow us
to track more extreme forms of housing instability that we cannot measure across all sites.
Data on eviction filings and judgments were collected from the Harris County Justice of
the Peace, the court entity that handles eviction cases, for calendar years 2019–2023. The
eviction records contain basic case details, such as filing date, amount owed, and disposition
or judgment details. We link these to the Harris County application data using names and
addresses as described in Appendix D.
Eviction court records for Cook County were shared by the Cook County Circuit Court
and supplemented with data from Record Information Services. Combined, the data span
the years 2019-2024. The data contain filing date, name, and address, and include filings that
were sealed from the public record as part of the state’s pandemic-related tenant protection
measures. Appendix D provides details on the linkage process. A limitation of both eviction
datasets is that we can link only to evictions occurring at the application address, which
undercounts total exposure to eviction.
3.5 Homelessness Data
We measure homeless services use in King County and Chicago using local Homeless Man-
agement Information System (HMIS) data. These databases systematically record the use
of homeless programs in a community, following the guidelines of the US Department of
Housing and Urban Development.
Starting from the set of all HMIS program enrollments between March 2019 and Septem-
ber 2021, King County staff extracted records that fuzzy matched with lottery applicants
based on name, date of birth, and ZIP code. We limit the data to program types offered
only to people who are already homeless (e.g., we do not include homelessness prevention
programs) so that program use indicates that the individual both receives services and is
homeless at the point of program entry.9 The existing literature uses similar measures to
9
The categories of programs we consider include emergency shelter, street outreach for people living
unsheltered, coordinated entry (entries for requesting but not yet accessing oversubscribed services), diversion
11
evaluate the impact of rental assistance and eviction on homelessness (Evans et al., 2016;
Phillips and Sullivan, 2024, 2025; Collinson et al., 2024).
The DOH and TRP applicant records were similarly matched on name and date of birth
to Chicago HMIS program enrollments spanning January 2017 to December 2023 through a
probabilistic linkage procedure. In the case of multiple viable matches to a single applicant,
HMIS service utilization histories were reconciled to take the highest level of engagement on
any given day. All outcomes for King County and Chicago were constructed to align with
HUD project types, except diversion services, which are only available in the King County
HMIS.
4 Who Were the Program Recipients?
4.1 Lottery Applicants
Table I provides summary statistics for each of the program sites. We report means separately
for applicants who were offered assistance (Z = 1) and those who were not (Z = 0). The table
summarizes applicant characteristics, which we group into three broad areas. The first panel
uses the complete lottery sample and provides information on demographics (or imputed
demographics) derived from the applications for assistance.10 The second panel reports ZIP
code characteristics of applicants from the American Community Survey. Finally, for those
who match to credit records, we report baseline credit measures from January 2020.
Applicants were, on average, 39 to 41 years old, and 54-64 percent of applicants were
female. The racial composition differs across lotteries. Black individuals account for over
one-half of applications in Harris County, and around a third in King County and Chicago
DOH. For Los Angeles and Chicago TRP, over 50 and 80 percent of applicants are Hispanic,
respectively. The median rent in applicants’ ZIP codes is between $1,000 and $1,500 across
sites, and poverty rates range from 11 to 21 percent. The average credit scores are low across
sites, ranging from 550 to 650, and applicants have relatively large balances in delinquency
or collections. Overall, the racial compositions differ across lotteries; otherwise, applicants
across sites are broadly similar.
Figure III extends the descriptive statistics from Table I to plot the evolution of several
key measures of financial distress: credit scores, total balances in delinquent accounts, and
total balances in collections for utility bills. Each panel reports the raw means (and standard
errors) for a given outcome and site separately for applicants who were and were not offered
programs provided to people who have just become homeless, transitional housing, permanent supportive
housing, and rapid rehousing.
10
For Harris County, we impute gender and race using the R packages gender, genderdata, and wru.
12
assistance.11 Means are plotted for all months with available credit reporting data. The
vertical dashed line denotes the month in which the lottery was conducted for each site. The
vertical red line marks the beginning of the pandemic (March 2020).12
Three features of this figure stand out. First, the balance between applicants who were
and were not offered assistance, which is documented in Table I for January 2020, persists
across all pretreatment months. The gap is somewhat larger in King County, where assis-
tance was randomly assigned but we only observe credit records for survey respondents, as
described in Section 3.1. Second, the overall trend for applicants across all sites is one of
generally improving financial distress. Credit scores trend upward throughout the sample
period, while delinquent balances and utilities collections trend downward. This finding is
consistent with evidence from other work showing that various measures of financial well-
being improved for most households during the early pandemic (Han et al., 2020). Third,
there is no meaningful divergence in outcomes between applicants who were and were not
offered assistance either at the onset of the pandemic or in the months following assistance.
This finding, which foreshadows our main results, is consistent with the idea that the emer-
gency assistance grants we study did not have meaningful effects on financial distress.
4.2 Did ERA Programs Target Those in Need?
As noted in Section 2, the ERA programs required proof of eligibility, including being below
income limits and an inability to pay rent or pandemic-related income loss. These criteria
may have led to the programs being accessed by lower-income or more financially strained
households compared to other pandemic-relief initiatives. To assess the extent to which
ERA programs targeted those in greater need, we first compare program recipients with
other renters in the same cities. Then, we compare ERA recipients to those who received
federal Economic Impact Payments (EIP), which were distributed around the same time.
Table II compares the characteristics of those who received assistance to the character-
istics of all renters who live within the relevant city or county. We compare demographic
characteristics (top panel), ZIP code characteristics (middle panel), and credit characteris-
tics (bottom panel). The first two panels use data on renters in the ACS in 2019.13 The final
11
We report standard errors for each mean separately rather than their difference to provide a sense of
the overall variability in the data. Formal tests for differences in means between applicants who were and
were not offered assistance are discussed in Section 6.
12
In King County, the applications were made and lotteries conducted over time. Thus, we display the
data by relative time, limiting to time periods for which data are available for the full sample. We indicate
the start of the pandemic using the average for the sample.
13
The first panel uses the 2019 ACS public-use micro-data and is a subset of all renters in each area.
The second panel uses aggregate ZIP code (ZCTA) characteristics from the 5-year 2015-2019 ACS, where
ZIP-level characteristics are weighted by the number of renters in the ZIP or the number of applicants in
13
panel uses data from a 10 percent random sample of individuals from Equifax. We restrict
to renters who live within one of the catchment areas for the programs in the months the
lotteries took place. We focus on credit scores (VantageScore 4.0), balance in delinquent
accounts, and balance in collections, as these outcomes can be best aligned between the
Equifax and Experian data.
Across all sites, those assisted are less likely to be White, and more likely to be Black
or Hispanic than renters on average. Those assisted are also from ZIP codes with higher
poverty rates, a lower share of White residents, and moderately lower median rent. Finally,
those assisted have notably lower credit scores, higher balances in collections or delinquency,
and are less likely to have a revolving line of credit (e.g., a credit card).
We further compare trends in the financial characteristics of those assisted compared
to all renters using the panel nature of the Equifax and Experian data. Appendix Figure
E.6 plots the outcome trajectories for applicants who were and were not offered assistance
and the random samples of renters over time relative to the treatment month for all five
programs. Across all sites, we see large gaps between the random sample of renters and
those who participated in the lottery. For example, in Harris and King Counties, the credit
score gap is more than 100 points in all periods, while the gap is 50 or more points in all
periods in Los Angeles and Chicago DOH. The lottery participants also have low average
baseline credit scores ranging between 550 and 640 (with the standard cut-off for subprime
being 600), with a smooth rise of approximately 25 points during the pandemic.
Similar gaps are present for the balances in delinquent accounts and in collections, with
those participating in the lottery having average balances in collections ranging from ap-
proximately $3,000 in Harris County and $1,400 to $1,800 in King County to $1,000 in Los
Angeles and $500 to $1,100 in Chicago. The trends are not widely different between lottery
participants and the random sample of renters, though on average, the gaps between the two
groups close somewhat over time. The random samples of renters have largely flat trajec-
tories, while both applicant groups have rising credit scores and flat or decreasing balances
over the whole time period.
Our second benchmark to evaluate the targeting properties of the ERA programs are the
recipients of the first round of federal Economic Impact Payments – cash checks provided
under the CARES Act. Figure II provides a comparison between recipients of the ERA
programs we study and EIP Round 1 recipients, based on ZIP code-level poverty rates.14
the ZIP.
14
We make comparisons based on ZIP code-level characteristics because we observe ZIP codes for all
recipients in the programs we study, and counts of EIP recipients by ZIP code are available from the Internal
Revenue Service Statistics of Income (Internal Revenue Service, 2020, 2021). We focus on EIP round 1
recipients because recipients in rounds 2 and 3 have nearly identical ZIP-level characteristics, as can be seen
14
Economic Impact Payments were the federal government’s most immediate response to the
economic impact of the pandemic and the first round consisted of checks in the amount of
$1,200 per adult plus $500 per qualifying child under age 17, sent to individuals with less
than $75,000 ($150,000 for married couples filing a joint return) in income reported on their
2018 or 2019 federal tax returns. Payment amounts were reduced for eligible individuals
with higher adjusted gross income. Across all sites, ERA recipients live in ZIP codes with
higher poverty rates as compared to EIP recipients.15
Appendix Figures E.1 and E.2 make similar comparisons based on ZIP-level minority and
Hispanic population shares for ERA recipients and EIP recipients. With the exception of Los
Angeles, all sites distributed assistance to recipients in neighborhoods with higher minority
shares. As expected given that it was designed to target undocumented immigrants, recipi-
ents in the Chicago TRP program were substantially more likely to be from neighborhoods
with high Hispanic shares. The differences between ERA and EIP recipient pools could be
the result of various factors, including the automatic transfer of the EIP payments versus the
opt-in procedure for ERA programs, documentation requirements for ERA versus selection
based on past tax returns for EIP, or the fact that ERA was more explicitly aimed towards
renters, while the EIP payments did not distinguish among homeowners and renters.
Overall, the evidence suggests that the ERA programs targeted renters who were substan-
tially more financially distressed, and who were living in higher-poverty and higher-minority
share neighborhoods, relative to renters not assisted by the programs. These findings are in
line with the results in Figinski et al. (2024), who find that the federal ERAP payments that
were implemented in a later phase of the pandemic (see Figure I) disproportionately went
to Census tracts with more disadvantaged renters.
5 Using Lotteries to Study the Impact of ERA
5.1 Regression Specifications
Our empirical analysis aims to estimate the causal effects of receiving ERA on various out-
comes of interest. However, simply comparing average outcomes across those who did and
did not receive assistance may yield biased estimates of this effect. Most financial assistance
programs select clients based on their vulnerability or their recent experience of a large
negative shock. Even in settings where assistance is administered via lottery, take-up of
in Appendix Figures E.3 and E.4.
15
Karger and Rajan (2021) provide evidence that those with lower income and less savings had higher
marginal propensities to consume from the EIP payments. This suggests that ERA may have targeted those
with higher marginal propensities to consume.
15
assistance is typically endogenous. Households that do not take up the assistance may differ
along dimensions unobservable to the econometrician. For example, they might have more
tumultuous lives, making it difficult to submit necessary paperwork or respond to attempts
to contact them. Alternatively, they might not take up the assistance because they have
relatively less need for it or have access to other financial support.
We address these challenges by estimating treatment effects using random variation in
offers of assistance. Specifically, we limit our sample to people who applied for assistance
and study contexts in which local jurisdictions rationed oversubscribed assistance by lottery.
To measure the effect of being offered assistance (which we will refer to as “intent-to-treat
effect,” or ITT), we estimate simple linear regressions of the following form:
Y = γ0 + γ1 Z + X′ ϕ + U, (1)
where Y is an outcome of interest, such as credit score, γ1 is the intent-to-treat effect, Z is
an indicator for whether the applicant was selected by lottery to receive an assistance offer,
and X is a set of control variables.16
As discussed in Appendix A, several jurisdictions administered their lotteries in a way
that resulted in offers that were random conditional on observable information, such as time
of application or Census tract. We include controls for these “site-specific design features” in
X in equation (1). In particular, we include week-of-application fixed effects for King County
and Census tract fixed effects for Harris County. For Chicago, we include application-count
fixed effects to account for a small number of individuals applying more than once. To
increase precision, we additionally control for individual-level gender and race indicators.
When analyzing credit report data, we also control for lagged credit outcomes for the period
prior to lottery application and for January 2020.17
Not everyone initially offered funding actually received it. There are several reasons those
selected for funding may not have received it, such as program administrators not being able
to contact the applicant, the applicant being determined ineligible, or the applicant failing
to complete the required forms. It is also possible that applicants not originally selected for
assistance via the lottery nonetheless ended up receiving it. To account for both directions
of imperfect take-up, our analysis focuses on instrumental variables estimates. We calculate
these estimates using two-stage least squares (TSLS) regressions, with the first and second
16
Below, we use the terminology “offered treatment” and “selected for an offer of treatment” interchange-
ably, even though not all applicants selected through the lottery were offered assistance, for example due to
difficulties contacting the applicant.
17
The lagged credit outcomes are credit score, total debt balance, balance in collections, balance delinquent,
utility balance in collections, and indicators for any delinquency, any revolving credit account, and any auto
loan.
16
stages given by:
R = α0 + α1 Z + X′ δ + V (2)
b + X′ λ + W,
Y = β0 + β1 R (3)
where R is an indicator variable for whether an applicant received assistance, and all other
variables are as previously defined. The coefficient β1 is our parameter of interest, and
represents a local average treatment effect (LATE) of assistance receipt if the standard IV
assumptions (Imbens and Angrist, 1994) are satisfied and (2)–(3) are correctly specified
(Blandhol et al., 2022).
Table III reports estimates of α1 from the first-stage regression for each site. The table
also reports the partial F-statistic on that coefficient, the control-group mean, and the share
offered assistance. For each site, we observe a large and statistically significant first stage
(with the F statistics ranging between 2,442 and 47,977). The size of the first-stage coef-
ficients differ as there were differential take-up rates across sites. Some of these differences
come from how the lotteries were implemented and, specifically, how much screening took
place at the time of the application versus upon selection as a lottery winner. For example,
Chicago’s initial application process was easy to complete, and candidates were screened
more rigorously upon being selected. Programs also differed in the extent to which assis-
tance required the agreement of landlords. In King County and during the early stages of
the Los Angeles program, the assistance was paid to landlords, some of whom did not accept
the program’s limits on future evictions or rent increases. In contrast, Harris County made
payments directly to the household applying for assistance.
We estimate all treatment effects site by site. For example, when measuring local av-
erage treatment effects, we estimate equations (2) and (3) separately for each site. We
then compute the simple, unweighted average of the estimates across sites. We estimate
heteroskedasticity-robust standard errors for each site and compute a full-sample standard
error assuming independence across sites.
5.2 Balance Tests
We check whether the data is consistent with randomization of offers of assistance by program
administrators. Table I shows the results of tests for balance on observable characteristics
between the groups offered and not offered assistance. For each site, we report the differ-
ence between these group means (“Diff”), adjusted using an ordinary least squares (OLS)
regression of equation (1) that controls for any site-specific design features (see Section 5.1).
Overall, the applicants selected for assistance are observably similar at baseline to appli-
17
cants not selected on nearly all variables. Across the demographic and credit characteristics,
the two groups are generally balanced. Where there are some statistically significant differ-
ences between the two groups’ populations, the differences are quantitatively small. Credit
report characteristics in King County show somewhat larger gaps between the two groups.
This is likely the result of the fact that we can only access Experian records in King County
for people who respond to the survey, as discussed in Section 3. This sample selection gen-
erates the somewhat larger baseline differences in attributes measured in credit reports, the
largest difference in credit scores across sites, 8.5 points. Reassuringly, the top two panels of
Table I use the full lottery sample for King County and show balance, suggesting that the
imbalance in Experian data is indeed due to selection into the survey in King County. Ap-
pendix Table E.1 reports balance for additional baseline applicant characteristics available
in Los Angeles, but not the other sites.
For most sites, survey response rates were not equal across the groups offered and not
offered assistance. To address potential bias due to differential selection into responding to
the survey, in any analysis of survey outcomes, we always weight the answers by the appli-
cant’s inverse probability of responding to the survey (Wooldridge, 2007). To calculate these
weights, we first estimate site-specific logit regressions of whether an applicant responded to
the survey, using a dummy for being offered assistance (Z) and a host of baseline control
variables. For all sites except King County, these controls include ZIP code-level median
rent, poverty rate, share high-school educated, share Hispanic, share White, share Black,
and share Asian. Additional applicant-level controls vary across sites depending on data
availability.18 We then predict an applicant’s probability of responding if they had not been
offered assistance (i.e., setting Z = 0) and use the inverse of this predicted probability as
the weight in our regressions. As discussed in Section 6.1, we obtain similar results with and
without these weights.
6 Results
In this section, we report estimates of treatment effects from receiving assistance on outcomes
measured in our surveys, financial distress measures observed in the administrative data
from Experian, residential mobility as measured in both the Infutor and Experian data, and
18
For Harris County, the controls include language (English speaking vs non-English speaking), contact
information (valid email, two phone numbers, etc.), public assistance type, income eligible, gender, race, and
ethnicity. For King County, they include indicators for race, gender, and primary language, and for both
Chicago lotteries, they include age and indicators for race, gender, missing age, and if the applicant applied
to the lottery more than once. For Los Angeles, these controls include age, household size, gender, language
(English speaking vs non-English speaking), ethnicity, race, income, monthly rent, total rent owed, months’
rent owed, and COVID hardship reason.
18
eviction and homelessness data as measured from court records and local homeless services
administrative systems. We report both site-specific estimates and equal-weighted averages
across sites.
6.1 Impacts on Survey Outcomes
Table IV reports IV estimates of the effect of assistance receipt on survey measures of rent
and bill payment, mental and physical health, material hardship, and feelings of economic
insecurity. Most of our survey questions ask about the month prior to survey collection.
However, the questions about rent payment ask about specific months following the month
of the lottery.19 Columns (1)–(4) report estimates separately by location, and column (5)
combines estimates across sites. Control group means (i.e., means for the group not selected
for assistance) are reported in brackets. While Table IV reports estimates from the weighted
regressions that adjust for survey nonresponse, the unweighted results are qualitatively sim-
ilar (Appendix Table E.3).
The survey data reveal three key results. First, assistance receipt led to higher rates of
rent payment at the time of follow-up. Consistently across Harris County, King County, and
Chicago (both DOH and TRP), recipients are 5.8–13.4 percentage points more likely to have
paid their rent in full in the month prior to the survey. Across these three sites, the combined
increase in the likelihood of rent payment is 8.2 percentage points, a 15 percent increase over
the control group mean. The effects of assistance on other types of bill payment appear more
mixed. Across the four lotteries, we find no evidence of impacts of the assistance on other
bill payments. Due to the substantive differences in the questions asked in the Los Angeles
survey, we report these estimates separately in Appendix Table E.4.20
Second, recipients are less likely to report feeling worried about being evicted (a 4.7
percentage-point decrease, or 16 percent). The point estimates for experiencing homelessness
(self-reported) are consistently negative, but none are statistically significant. The combined
estimate for staying in a homeless shelter is not statistically significant overall. However, we
do find a statistically significant decrease in self-reported sheltered homelessness in Harris
County (where the CDC moratorium was not as stringently enforced). The other sites have
a mixture of negative and positive point estimates which are not significant.21 We also find
no statistically significant evidence of reduced food insecurity.
19
See Appendix B for details about the survey questions across each site.
20
The effects are noticeably different for Los Angeles, likely due to difference in the wording of the survey
question, which focused on cumulative rent payment. At the same time, there is some indication in LA
that receipt of assistance resulted in increased payment of bills other than rent, though the estimates are
somewhat imprecise.
21
We consider impacts on homelessness as measured by administrative records in Section 6.4.
19
Finally, there is suggestive evidence of improved physical and mental health among re-
cipients. Recipients in the Chicago DOH lottery experienced a 5.1 percentage-point decline
in their likelihood of becoming sick in the previous month as a consequence of COVID-19,
and the point estimates for King County and Harris County are of similar magnitude, albeit
measured with less precision. Across all three sites, we estimate a marginally significant 5.6
percentage point reduction. Similarly, recipients are less likely to have reported feelings of
anxiety or depression, a 3.2-percentage-point (6.3 percent) decrease across Harris County,
King County, and Chicago, which is marginally significant.
Overall, the results based on the survey data suggest that recipients were more likely
to pay rent and to be less worried about being evicted. The point estimates for staying
in shelter also suggest reductions in housing instability, but are generally not significant.
Lastly, we find suggestive evidence that assistance modestly improved recipients’ physical
and mental health.
6.2 Impacts on Credit Measures
Next, we consider how financial assistance affected financial health using panel data from
Experian. Figure IV depicts reduced-form (ITT) and IV estimates, constructed as described
in Section 5.1, for a subset of our credit outcomes. Each column reports effects on a different
outcome: credit score (left column), balance in delinquent accounts (middle column), and
balance in utilities collections (right column). The reduced-form (ITT) and IV estimates are
depicted by navy diamonds and gold circles, respectively. Across outcomes and locations,
there is a consistent pattern in the results—namely, that being selected to receive assistance
(ITT) or receiving assistance (IV) did not meaningfully improve applicants’ financial well-
being.
Table V summarizes the IV estimates of the effects of assistance across a broad range of
credit outcomes measured two months and ten months after receipt. There is some evidence
that recipients used the additional resources to pay down some of their debt in the short
run. Receipt of assistance led to a $74 decrease in the debt balance in collections (column
(6)), driven by declines in the King County and Los Angeles samples. However, there are
no consistent effects on credit score, total balances, balance in utility collections, delinquent
accounts, or bankruptcy. Recipient households are no more likely to have increased their
durable consumption by taking out a car loan/lease and no more likely to have used a
revolving line of credit. Within ten months of assistance receipt, there are no statistically
significant differences in the administrative credit outcomes.
The overall effects on credit outcomes are small and precisely estimated. For example, the
20
95 percent confidence interval for the short-run effect on credit scores ranges from -3.1 to 1.8.
Thus, we can rule out large effects on credit scores like the 16.5-point decrease in response
to an eviction estimated in Collinson et al. (2024) or the 9.4-point increase in response to
the removal of a bankruptcy flag estimated in Dobbie et al. (2020). Similarly, the 95 percent
confidence interval on the 10-month balance in collections, which ranges from -$175 to $196,
allows us to rule out effects of the magnitude of the $302 increase in collections balances
4 years after a hospitalization among nonelderly insured adults estimated in Dobkin et al.
(2018).
6.3 Impacts on Residential Mobility
We find no consistent evidence that residential mobility changed as a result of receiving
emergency rental assistance. Table VI reports our IV estimates of the effects of assistance
on residential mobility. We measure address changes in two administrative data sets in
the short run (two months post-lottery) and long run (ten months post-lottery). Using the
Infutor data, we code a change of address if an applicant’s current address differs from her
address at the time of application. In the Experian data, address changes are measured by
changes in ZIP code relative to the application ZIP code. Move rates do not differ noticeably
with receipt of assistance. For example, 10-month move rates in the Infutor data decrease
by a statistically insignificant 0.8 percentage points from a mean of 9.3, and the 95 percent
confidence interval on this estimate ranges from -2.7 to 1.2 percentage points. For context,
an eviction order is estimated to increase the likelihood of a changed address one year later
by 8.2 percentage points (Collinson et al., 2024).
6.4 Impacts on Eviction and Homelessness
Next, we investigate whether the assistance reduced other measures of acute housing insta-
bility, including evictions and homelessness. We are not able to track these outcomes across
all sites, so we draw on evidence from individual locations. In particular, we study effects
on homelessness in Chicago and King County and effects on eviction activity in Chicago and
Harris County.
In Table VII, we report the effects of receiving assistance on homelessness system use
in the 9 months after the lottery for three of the lotteries that we study: King County,
Chicago DOH, and Chicago TRP. We examine the impacts on any homelessness system
use as in Collinson et al. (2024) and Phillips and Sullivan (2025) and the impacts on the
types of services used. The effects of assistance on homelessness are mixed. In King County,
receipt of assistance increased interactions with the homelessness system by 2 percentage
21
points. Descriptively, this result appears to be driven primarily by unassisted members
of the assigned treatment group being made more likely to show up in the homelessness
system, as we explore in Appendix Figure E.7. This finding could be due to applicants
whose applications were incomplete or who were found to be ineligible being steered to other
resources in the homelessness system by program staff.
In Chicago, applicants who received assistance through the TRP lottery were 0.35 per-
centage points less likely to appear in the homelessness system in the 9 months after the
lottery. This decline is a small absolute reduction but a large relative reduction of 65 percent
since homelessness system interactions are rare among the TRP applicants not selected for
assistance via lottery (only 0.55 percent of control group applicants end up in the home-
lessness system). Applicants who received assistance through the Chicago DOH lottery are
slightly less likely to have interacted with the homelessness system, but the difference is not
statistically significant.
Our results on homelessness from King County and Chicago DOH contrast with other
evidence on emergency rental assistance programs from outside the pandemic context. Prior
work finds that emergency rental assistance decreases rates of homelessness by approximately
three-quarters in Chicago (Evans et al., 2016) and San Jose (Phillips and Sullivan, 2025).
This contrast between our results and others from the literature does not arise from major
differences in the interventions being studied. The emergency financial assistance programs
offered during COVID were largely designed based on existing programs and in most cases
were even operated by the same agencies. The traditional and COVID-era programs both
feature temporary assistance in relatively modest amounts. For example, in Evans et al.
(2016), Phillips and Sullivan (2025), and Phillips and Sullivan (2024), the clients offered
financial assistance received an average of 0.7, 1.0, and 0.7 months of rent, respectively,
which are magnitudes similar to those of the programs that we study in Chicago, Harris
County, and Los Angeles, though the King County program is somewhat more generous.
The programs that we study also target clients who appear largely similar to the par-
ticipants in pre-COVID-era rental assistance programs. The King County COVID rental
assistance lottery and the pre-COVID homelessness prevention program studied in Phillips
and Sullivan (2024) were operated by the same agency, are evaluated using the same data
sources and methods, and had enrollment periods separated by only 5 months, providing
an ideal comparison. In the King County COVID program, 19.7 percent of the sample had
moved in the 12 months prior to the onset of the pandemic in March 2020, compared to 18.1
percent for the sample in Phillips and Sullivan (2024). Homelessness service usage differs
slightly between the two samples, at 4.8 and 7.9 percent, respectively, but both are high.
Similarly, the DOH lottery population in Chicago has rates of prior homelessness compa-
22
rable to those studied in (Evans et al., 2016). An exception is the Chicago TRP lottery,
which served a population with much lower pre-COVID rates of homelessness system use
than those of the populations studied in prior work.
In contrast to King County and Chicago DOH, our results on homelessness for the
Chicago TRP lottery are qualitatively similar to the findings from prior studies of pre-COVID
emergency rental assistance programs (Evans et al., 2016; Phillips and Sullivan, 2025), show-
ing a sharper reduction in homelessness for those who received assistance through this lottery
than for those treated by the other lotteries that we evaluate here. A plausible explanation
for this difference is that the TRP program targeted populations who might have more lim-
ited access to other aspects of the pandemic safety net, such as expanded unemployment
insurance or COVID Economic Impact Payments, either because they were undocumented
immigrants or lacked formal sector employment. In Appendix Table E.12, we report the
fraction of applicants receiving government assistance across lotteries. Consistent with pro-
grammatic differences in target populations, we find that TRP applicants report much lower
rates of assistance from government sources in our survey than Chicago DOH applicants or
applicants in King County.
Finally, to examine the impacts on eviction filing, we link the sample of Harris County
and Chicago assistance applicants to eviction court records.22 As detailed in Appendix
A, the interpretation of the CDC moratorium varied by local courts, as did the presence
of extended state or local moratoriums. In Harris County, judges frequently permitted
eviction non-payment cases to proceed during the federal moratorium when tenants failed
to file a Covid hardship declaration form. Additionally, cases for reasons other than non-
payment were generally permitted to continue as usual (Schuetz, 2021). However, formal
enforcement of eviction orders by law enforcement was stayed during the moratorium. In
Chicago, adherence to the CDC moratorium was stronger, and both a statewide moratorium
and a city ordinance extended protections several months beyond the CDC moratorium.
Table VIII reports the effects of receipt of assistance on eviction filing (initiation of a new
case), which we observe in Harris County and Chicago (both DOH and TRP). One limitation
of this analysis is that we can only consider evictions occurring at the same address listed
on the application for assistance. Due to the aforementioned differences across courts in
eviction activity during this period, and the variation over time in the presence of the federal
moratorium, we focus on evaluating the effects of assistance separately by lottery, as well as
during the CDC moratorium and after it was lifted. The top panel considers 2SLS on any
eviction filing within two, six, and ten months after the lottery (i.e. during the moratorium).
The bottom panel focuses on impacts on eviction filings two, six, and ten months after the
22
This linkage is described in Appendix D.
23
CDC moratorium was lifted. The first column for each lottery reports the control mean. As
expected, the baseline means during the moratorium are quite low, especially in Chicago.23
After the moratorium, baseline means remain low, potentially partly due to applicants no
longer living at the same address, but are several times higher than during the moratorium.
The second column for each site in Table VIII reports the IV estimates for the effects
of the assistance. For Harris County, we find no statistically significant effects, with small
negative point estimates in all periods. For Chicago DOH, we find some suggestive evidence
of reduced evictions. During the moratorium, we find a reduction two months after the
lottery. Yet, the control mean (0.0004) and treatment effect (-0.0004) are so small that this
is based only on an extremely small difference in filings and should be interpreted cautiously.
After the moratorium the estimates for Chicago DOH are also negative, with a weakly
statistically significant reduction of -0.003 after two months, off a baseline mean of 0.005.
The estimates for Chicago TRP are all statistically insignificant and small. In summary,
the results are consistent with a small reduction in eviction filings during and after the
moratorium, but the estimates are quantitatively small.
Taking the above results together, we find little evidence that applicants who received
emergency rental assistance through the lottery were less likely than nonselected applicants
to experience extreme housing instability in the months after applying. These results are
consistent with our finding of limited to no effects of assistance on residential mobility or
credit market outcomes.
6.5 Heterogeneity and Mechanisms
The modest overall effects may mask larger effects across some subgroups. Mechanisms that
might drive those modest results (such as the presence of other generous federal benefits, a
weak housing market that allows for negotiation with landlords, or eviction moratoria) also
imply possible heterogeneous effects. Here, we test for heterogeneity across both subgroups
and sites.
We find little subgroup heterogeneity in outcomes measured by credit reports. Appendix
Figure F.1 shows credit report outcomes, dividing subgroups based on both levels and trends
in credit measures, including groups with credit scores below (above) 580, delinquent account
balances above (below) $5,000, decreasing (increasing) credit scores, and increasing (decreas-
ing) delinquent account balances. All these figures plot standardized IV estimates for the
23
In Panel D of Figure I we plot monthly eviction filings for all four sites. In Harris County, filing volumes
drop with the onset of COVID-19 and the CARES Act Moratorium. Filings remain about one-third of
pre-pandemic levels (though still average about 2,000 cases per month) through the CDC moratorium. In
contrast, the number of filings in Chicago (and King County) remain lower during the CDC moratorium.
24
effects of assistance.24 Overall, there are few meaningful differences between the subgroups.
We also find little heterogeneity in the survey results. While we lack baseline surveys,
we use retrospective survey questions about pre-lottery conditions to construct 12 subgroups
capturing different economic circumstances, housing situations, and health conditions. In
particular, we examine effects separately by the following pre-lottery subgroups: high-/low-
income, some/no safety-net utilization, good/poor mental health, stable/unstable employ-
ment, current/behind on rent, and moved vs not in the past year. Due to the variation in
the sample size of some subgroups between sites, we inverse-variance weight the estimates
from individual locations when combining point estimates across sites.25 The results of the
heterogeneity of the survey appear in table form in Appendix Tables F.2 and F.1, and as
forest plots in Appendix Figures F.2 and F.3. The effects on rent payment and bill payment
in Appendix Figure F.1 appear to vary relatively little across subgroups. There are slightly
larger point estimates for rent payment among those not stably employed and those those
who report being behind on their rent in the past. The health effects also appear mostly
stable across subgroup. We find some evidence that reductions in anxious feelings are larger
for individuals who report more economic precarity before the lottery, particularly those
who report being behind on rent before the lottery, and those not stably employed. Overall,
however, we observe little heterogeneity.
One possible exception to this lack of individual-level, within-site, heterogeneity is pre-
lottery, self-reported safety net utilization. This particular sample split is of interest because
people with limited pre-lottery experience with the safety net probably have less access
to substitutes for ERA.26 In Appendix Table F.2 we report the impacts of assistance on
economic security measures from Table IV separately for those who report some pre-lottery
safety-net utilization versus those who report no safety-net program receipt. We consistently
find larger point estimates for the applicants who report not receiving other safety-net help
prior to the lottery. For example, those with no reported baseline safety-net access experience
a 3.9 percentage point reduction in reported homelessness, which is statistically significant.
Meanwhile, the effect for those reporting pre-existing safety-net access is an estimated zero.
Effects appear similarly larger for worries about eviction, stays in shelters, and food insecu-
rity, though in most cases we are not able to formally reject equality. Overall, while only
suggestive, this evidence is consistent with the idea that ERA was more effective among
24
We standardize the IV estimates by dividing the point estimates by the control group standard deviation
of the outcome.
25
We show in Appendix Tables F.3 and F.4 that the point estimates are quite similar to the equally-
weighted estimates, but are considerably more precise.
26
While there are conceptual reasons to think this group might have different effects, we note that this
was not a pre-registered subgroup, so we encourage caution in interpreting the results.
25
people with less access to other parts of the safety net.
One strength of this study is the ability to observe treatment effects of ERA across
four metropolitan areas that varied in how they implemented both ERA and other relevant
pandemic-era policies, like eviction moratoria. Some cross-site results, like stronger effects
on housing stability among primarily undocumented applicants in the Chicago TRP lottery,
suggest that such comparisons may be informative about mechanisms. So, in addition to
heterogeneity on observables, we formally test for heterogeneous treatment effects across
sites. Using tools from the meta-analysis literature, we consider a random-effects model
where site-specific estimates may differ from one another due to sampling variability and
differences in the true impacts across sites (DerSimonian and Laird, 1986; Jackson and
Mackevicius, 2024). In Appendix F.2, we describe the implementation and present the
results. Overall, we find limited evidence of cross-site heterogeneity. The strongest evidence
is for financial outcomes 10 months later, where we reject the null of homogeneity across sites
for credit scores and for an indicator for any personal bankruptcy. Yet, overall, we rarely
reject the null of no site-level heterogeneity. These null results are consistent with the idea
that the modest effects of ERA that we measure are due to national context shared by all
sites, like the availability of other federal assistance or a slack housing market, rather than
features that vary more across sites, like ERA program design or the strictness of eviction
moratoria. However, with only five lotteries and four locations, these tests have limited
statistical power.
7 Discussion and Concluding Remarks
This paper studies the unprecedented expansion of emergency rental assistance in response
to the economic disruption caused by the COVID-19 pandemic. Using application data from
five program lotteries conducted in four large cities linked to credit records, administrative
homeless system data, and novel surveys, we provide the first experimental evidence on the
effects of pandemic-era ERA on a wide range of policy-relevant outcomes.
We show that these programs were relatively well-targeted. Compared to other renters
in the area, applicants were more financially distressed, more likely to live in high-poverty
neighborhoods, and more likely to be minorities. Similar conclusions are reached when
comparing applicants to EIP recipients. Lastly, the assistance went out relatively early in
the pandemic while unemployment rates and rental arrears were still quite high, and well
before the federal ERA programs.
We find that rental assistance reduced rental arrears and concern about eviction, and
may have improved self-reported physical and mental health. However, it had little de-
26
tectable short- or longer-run effect on broader measures of housing stability and financial
security, including evictions, homelessness, residential mobility, credit scores, indebtedness,
or delinquency.
The key question raised by these findings is: why did these programs not have more of
a detectable impact? Several pieces of evidence suggest that the modest effects that we see
are unlikely to be a result of poor targeting or insufficient support. The program applicants
were substantially more financially distressed than the average renter in their communities,
and the null effects that we find persist even in subsamples of applicants with severe financial
distress at the time of application. The size of the grants are also similar to that of other
smaller-scale assistance programs that have been shown to have large effects (Phillips and
Sullivan, 2025; Evans et al., 2016). Moreover, we find little heterogeneity across sites, even
though the assistance was significantly more generous in some places (e.g., King County).
These results suggest that the broader economic and policy context, rather than program
design or targeting, was responsible for the modest size of the effects. Unlike other ERA
programs, which target shocks faced by individual renters, the programs that we study were
created with the specific aim of responding to the COVID pandemic, which was an aggregate
shock affecting all households. Below, we discuss three natural explanations for why ERA
could be less impactful in this context.
First, aggregate shocks tend to be accompanied by other fiscal stimulus that provides
additional support to low-income households, which tend to be renters. The COVID-19
pandemic, in particular, featured a massive expansion of nearly every aspect of the social
safety net, including Unemployment Insurance, Economic Impact Payments, and expansions
of the Earned Income Tax Credit and Child Tax Credit. Consistent with this explanation,
we find that individuals who applied for assistance—regardless of whether they actually
received it—experienced broad-based improvements in their financial circumstances in the
post-application period. The improvement in household balance sheets continues post-lottery
but is similar for both treatment and control applicants. Furthermore, we find evidence that
applicants with less access to other safety net alternatives appear to benefit more from ERA
receipt. Together, this suggests that the widespread availability of other forms of assistance
reduced the marginal impact of ERA.
Second, eviction moratoria likely weakened the link between rental arrears and displace-
ment. While moratoria did not eliminate housing instability—landlords could still pursue
evictions for lease violations, decline to renew expiring leases, or pressure tenants through
informal means—they nonetheless reduced the immediate risk of court-ordered displacement
following nonpayment. This disruption to the typical mechanism through which ERA pre-
serves housing stability likely reduced the program’s effectiveness. The threat of future
27
eviction action remained, which likely explains why we observe ERA reducing tenant anxi-
ety about eviction. However, the temporary legal protection from displacement meant that
ERA’s marginal contribution to preventing actual moves was limited during the period when
moratoria were in effect.
A third explanation is that general equilibrium market responses to aggregate shocks may
limit the impact of additional rental assistance, even in the absence of moratoria or expanded
fiscal stimulus. During the beginning of the COVID pandemic, landlords faced both falling
rents and rising vacancies, creating incentives to retain tenants, even those with arrears.
The prospect of extended vacancies and uncertain rental income could make both immediate
eviction and lease non-renewal less attractive options. This shift in market dynamics could
reduce displacement risk regardless of formal assistance, further weakening the connection
between ERA and housing stability outcomes. Thus, renegotiation or temporary agreements
between landlords and tenants may be sufficient to forestall eviction without additional
emergency assistance. Consistent with this possibility, we find that more than 70 percent of
the lottery applicants who missed rental payments reported having some type of agreement
with their landlord (Appendix Table E.11). This evidence is in line with other pandemic-
era surveys of landlords that find substantial increases in reported rental forbearance in
2020, relative to pre-pandemic levels (Decker, 2021; de la Campa and Reina, 2023). In this
context, rental assistance may reduce rental arrears without changing tenants’ longer-run
housing stability, because these arrears would not have led to eviction or homelessness, even
in the absence of assistance.
While extensive evidence from the Great Recession has yielded lessons on the effective-
ness of foreclosure prevention policies, far less is known about crisis-response policies for
renters. Our lottery-based estimates of the impact of local ERA programs implemented
during the COVID-19 pandemic allow us to measure the effects of one such policy dur-
ing an unprecedented combination of economic disruption and policy intervention. Our
results suggest that ERA’s effectiveness may depend critically on the broader policy en-
vironment: Programs that substantially reduced homelessness in normal times had muted
effects when deployed alongside eviction moratoria and expanded safety net programs. This
apparent context-dependence raises important considerations for future crisis response, par-
ticularly regarding how and when to deploy rental assistance relative to other emergency
measures. Whether our results generalize to later rounds of ERA—which were implemented
after the pandemic’s initial impact, were typically more generous, and prioritized paying
down arrears—remains an open question. Future work should examine both the longer-term
effects of these programs and their impacts on landlord behavior and rental market dynamics.
28
References
Agarwal, Sumit, Gene Amromin, Itzhak Ben-David, Souphala Chomsisengphet,
Tomasz Piskorski, and Amit Seru, “Policy Intervention in Debt Renegotiation: Evidence
from the Home Affordable Modification Program,” Journal of Political Economy, June 2017, 125
(3), 654–712.
, , Souphala Chomsisengphet, Tim Landvoigt, Tomasz Piskorski, Amit Seru,
and Vincent Yao, “Mortgage Refinancing, Consumer Spending, and Competition: Evidence
from the Home Affordable Refinance Program,” The Review of Economic Studies, June 2022,
p. rdac039.
Airgood-Obrycki, Whitney, “The Short-Term Benefits of Emergency Rental Assistance,” Tech-
nical Report, Joint Center for Housing Studies of Harvard University June 2022.
An, Xudong, Stuart A. Gabriel, and Nitzan Tzur-Ilan, “More Than Shelter: The Effects
of Rental Eviction Moratoria on Household Well-Being,” September 2021.
Asquith, Brian J., Evan Mast, and Davin Reed, “Local Effects of Large New Apartment
Buildings in Low-Income Areas,” The Review of Economics and Statistics, January 2023, pp. 1–
17.
Benfer, Emily A., Anne Kat Alexander, Alieza Durana, Robert Koehler, and Matthew
Desmond, “COVID-19 Housing Policy Scorecard,” 2020. Place: Princeton: Princeton Univer-
sity.
Blanchet, Thomas, Emmanuel Saez, and Gabriel Zucman, “Real-Time Inequality,” Tech-
nical Report w30229, National Bureau of Economic Research, Cambridge, MA July 2022.
Blandhol, Christine, John Bonney, Magne Mogstad, and Alexander Torgovitsky,
“When is TSLS Actually LATE?,” Technical Report, National Bureau of Economic Research
2022.
Chetty, Raj, John N Friedman, Michael Stepner, and Opportunity Insights Team,
“The Economic Impacts of Covid-19: Evidence from a New Public Database Built Using Private
Sector Data*,” The Quarterly Journal of Economics, October 2023, p. qjad048.
Collinson, Robert, Ingrid Gould Ellen, and Jens Ludwig, “Reforming Housing Assistance,”
The ANNALS of the American Academy of Political and Social Science, November 2019, 686
(1), 250–285.
, John Eric Humphries, Nicholas Mader, Davin Reed, Daniel Tannenbaum, and Win-
nie van Dijk, “Eviction and Poverty in American Cities,” The Quarterly Journal of Economics,
February 2024, 139 (1), 57–120.
Collyer, Sophie, Jill Gandhi, Irwin Garfinkel, Schuyler Ross, Jane Waldfogel, and
Christopher Wimer, “The Effects of the 2021 Monthly Child Tax Credit on Child and Fam-
ily Well-Being: Evidence from New York City,” Socius, January 2022, 8, 23780231221141165.
Publisher: SAGE Publications.
de la Campa, Elijah A. and Vincent J. Reina, “Landlords’ rental businesses before and after
the COVID-19 pandemic: Evidence from a National Cross-Site Survey,” Journal of Housing
Economics, March 2023, 59, 101904.
Decker, Nathaniel, “The Uneven Impact of the Pandemic on the Tenants and Owners of Small
Rental Properties,” Technical Report, Terner Center for Housing Innovation UC Berkeley 2021.
29
DerSimonian, Rebecca and Nan M. Laird, “Meta-analysis in clinical trials,” Controlled Clin-
ical Trials, September 1986, 7 (3), 177–188.
Diamond, Rebecca, Tim McQuade, and Franklin Qian, “The Effects of Rent Control Expan-
sion on Tenants, Landlords, and Inequality: Evidence from San Francisco,” American Economic
Review, September 2019, 109 (9), 3365–3394.
Dobbie, Will, Paul Goldsmith-Pinkham, Neale Mahoney, and Jae Song, “Bad Credit,
No Problem? Credit and Labor Market Consequences of Bad Credit Reports,” The Journal of
Finance, 2020, 75 (2), 2377–2419.
Dobkin, Carlos, Amy Finkelstein, Raymond Kluender, and Matthew J. Notowidigdo,
“The Economic Consequences of Hospital Admissions,” American Economic Review, 2018, 108
(2), 308–352.
Evans, William N., James X. Sullivan, and Melanie Wallskog, “The impact of homelessness
prevention programs on homelessness,” Science, August 2016, 353 (6300), 694–699.
Fiano, Liane, “Protecting your credit during the coronavirus pandemic,” July 2020.
Figinski, Theodore F., Sydney Keenan, Richard Sweeney, and Erin Troland, “Targeted
Relief: Geography and Timing of Emergency Rental Assistance,” 2024.
Fulford, S., The Pandemic Paradox: How the COVID Crisis Made Americans More Financially
Secure, Princeton University Press, 2023.
Ganong, Peter and Pascal Noel, “Liquidity versus Wealth in Household Debt Obligations:
Evidence from Housing Policy in the Great Recession,” American Economic Review, October
2020, 110 (10), 3100–3138.
, Fiona Greig, Pascal Noel, Daniel Sullivan, and Joseph Vavra, “Spending and Job-
Finding Impacts of Expanded Unemployment Benefits: Evidence from Administrative Micro
Data,” Technical Report w30315, National Bureau of Economic Research, Cambridge, MA Au-
gust 2022.
Glasner, Benjamin, Oscar Jiménez-Solomon, Sophie M. Collyer, Irwin Garfinkel, and
Christopher T. Wimer, “No Evidence The Child Tax Credit Expansion Had An Effect On The
Well-Being And Mental Health Of Parents,” Health Affairs, November 2022, 41 (11), 1607–1615.
Publisher: Health Affairs.
Han, Jeehoon, Bruce D. Meyer, and James X. Sullivan, “Income and Poverty in the COVID-
19 Pandemic,” Brookings Papers on Economic Activity, 2020, 2020 (2), 85–118. Publisher: Johns
Hopkins University Press.
Hepburn, Peter, Renee Louis, and Matthew Desmond, “Eviction Tracking System: Version
1.0,” 2020. Place: Princeton: Princeton University.
Household Debit and Credit Report
Household Debit and Credit Report, Technical Report, Federal Reserve Bank of New York
November 2023.
Imbens, Guido and Josh Angrist, “Identification and Estimation of Local Average Treatment
Effects,” Econometrica, 1994, 62 (2), 467–475.
Internal Revenue Service, “Statistics of Income: Economic Impact Payments Data,” 2020.
accessed August 10, 2024.
30
, “Statistics of Income: Economic Impact Payments Data,” 2021. accessed August 20, 2024.
Jackson, C. Kirabo, “When Decentralization Works: Leadership, Local Needs, and Student
Achievement,” Working Paper 34216, National Bureau of Economic Research September 2025.
and Claire L. Mackevicius, “What Impacts Can We Expect from School Spending Policy?
Evidence from Evaluations in the United States,” American Economic Journal: Applied Eco-
nomics, January 2024, 16 (1), 412–46.
Jaroszewicz, Ania, Jon Jachimowicz, Oliver Hauser, and Julian Jamison, “How Effective
Is (More) Money? Randomizing Unconditional Cash Transfer Amounts in the US,” Technical
Report, SSRN Scholarly Paper No. 4154000, Rochester, NY July 2022.
Jowers, Kay, Christopher Timmins, Nrupen Bhavsar, Qihui Hu, and Julia Marshall,
“Housing Precarity & the COVID-19 Pandemic: Impacts of Utility Disconnection and Evic-
tion Moratoria on Infections and Deaths Across US Counties,” Working Paper 28394, National
Bureau of Economic Research, Cambridge, MA January 2021.
Karger, Ezra and Aastha Rajan, “Heterogeneity in the Marginal Propensity to Consume:
Evidence from Covid-19 Stimulus Pay,” Technical Report WP 2020-15, Federal Reserve Bank of
Chicago February 2021.
Kovski, Nicole, Natasha V. Pilkauskas, Katherine Michelmore, and H. Luke Shaefer,
“Unconditional cash transfers and mental health symptoms among parents with low incomes:
Evidence from the 2021 child tax credit,” Population Health, June 2023, 22, 101420.
McCarty, Maggie and Libby Perl, “Federal Eviction Moratoriums in Response to the COVID-
19 Pandemic,” Technical Report IN11516, Congressional Research Service 2021.
, David H. Carpenter, and Libby Perl, “The CDC’s Federal Eviction Moratorium,” Tech-
nical Report IN11673, Congressional Research Service 2021.
Phillips, David C., “Measuring Housing Stability With Consumer Reference Data,” Demography,
June 2020, 57, 1323–1344.
Phillips, David C and James X Sullivan, “Personalizing homelessness prevention: Evidence
from a randomized controlled trial,” Journal of Policy Analysis and Management, 2024, 43 (4),
1101–1128.
and , “Do homelessness prevention programs prevent homelessness? Evidence from a ran-
domized controlled trial,” Review of Economics and Statistics, 2025, 107 (5), 1187–1196.
Pignatti, Clemente and Zachary Parolin, “The Effects of an Unconditional Cash Transfer on
Mental Health in the United States,” June 2023.
Pilkauskas, Natasha V, Brian Jacob, Elizabeth Rhodes, Katherine Richard, and Luke
Shaefer, “The COVID Cash Transfer Study: The Impacts of Unconditional Cash Transfers on
Low-Income Families,” Technical Report, Working Paper May 2022.
Pilkauskas, Natasha V., Katherine Michelmore, and Nicole Kovski, “The 2021 Child Tax
Credit, the Living Arrangements and Housing Affordability of Families with Low Incomes,” June
2023.
Piskorski, Tomasz and Amit Seru, “Mortgage Market Design: Lessons from the Great Reces-
sion,” Brookings Papers on Economic Activity, 2018, 2018 (1), 429–513.
31
Rafkin, Charlie and Evan Soltas, “Eviction as Bargaining Failure: Hostility and Misperceptions
in the Rental Housing Market,” Working Paper, March 2024.
Raphael, Steven and Daniel Schneider, “Introduction: The Socioeconomic Impacts of COVID-
19,” RSF: The Russell Sage Foundation Journal of the Social Sciences, May 2023, 9 (3), 1–30.
Publisher: RSF: The Russell Sage Foundation Journal of the Social Sciences.
Reina, Vincent, Claudia Aiken, Julia Verbrugge, Ingrid Gould Ellen, Tyler Haupert,
Andrew Aurand, and Rebecca Yae, “COVID-19 Emergency Rental Assistance: Analysis
of a National Survey of Programs,” Technical Report, National Low Income Housing Coalition
January 2021.
Reina, Vincent J. and Yeonhwa Lee, “COVID-19 and Emergency Rental Assistance: Impact
on Rent Arrears, Debt, and the Well-Being of Renters in Philadelphia,” RSF: The Russell Sage
Foundation Journal of the Social Sciences, May 2023, 9 (3), 208–229. Publisher: RSF: The
Russell Sage Foundation Journal of the Social Sciences.
Schuetz, R.A., “In Harris County, hundreds evicted despite moratorium,” Houston Chronicle,
2021.
Stein, Sarah and Nisha Sutaria, “Housing Policy Impact: Federal Eviction Protection Coverage
and the Need for Better Data,” Technical Report, Atlanta Federal Reserve 2020.
Sudeall, Lauren, Elora L. Raymond, and Philip M.E. Garboden, “Disaster Discordance:
Local Court Implementation of State and Federal Eviction Prevention Policies during COVID-
19,” Georgetown Journal on Poverty Law & Policy, 2023.
U.S. Census Bureau, “2020 Census Demographic and Housing Characteristics File,” 2020. ac-
cessed August 20, 2024.
, “American Community Survey (ACS) 5-Year Data (2020) Subject Tables,” 2020. accessed
August 20, 2024.
, “Week 13 Household Pulse Survey: August 19- August 31,” Technical Report, United States
Census Bureau 2020.
Wooldridge, Jeffrey M., “Inverse probability weighted estimation for general missing data prob-
lems,” Journal of Econometrics, December 2007, 141 (2), 1281–1301.
Yae, Rebecca, Sophie Siebach-Glover, Andrew Aurand, and Meredith Abel, “Continu-
ing Emergency Rental Assistance: How Jurisdictions Are Building on Treasury’s ERA Program,”
July 2023.
32
Tables and Figures
33
TABLE I
Summary Statistics and Balance
Harris County King County Los Angeles Chicago DOH Chicago TRP
Z=0 Z=1 Diff Z=0 Z=1 Diff Z=0 Z=1 Diff Z=0 Z=1 Diff Z=0 Z=1 Diff
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15)
Demographics
Age − − − 38.729 38.528 0.098 41.577 41.315 −0.262 38.367 38.741 0.374 38.991 38.944 −0.046
Female 0.648 0.642 0.013 0.618 0.618 0.005 0.546 0.541 −0.005 0.610 0.592 −0.019 0.562 0.560 −0.002
Household Size − − − − − − 2.660 2.664 0.005 2.990 3.002 0.012 4.050 4.004 −0.047
White 0.061 0.051 −0.003 0.231 0.235 0.007 0.213 0.217 0.003 0.218 0.223 0.004 0.039 0.034 −0.006
Black 0.607 0.629 −0.007 0.325 0.345 −0.027* 0.116 0.115 −0.001 0.333 0.336 0.003 0.120 0.112 −0.008
Hispanic 0.259 0.253 0.015** 0.177 0.153 0.008 0.535 0.540 0.005 0.406 0.390 −0.016 0.800 0.821 0.021*
Asian 0.024 0.025 −0.007***0.074 0.073 −0.001 0.109 0.105 −0.004 0.043 0.052 0.009* 0.041 0.032 −0.009
ACS Zip Code Data
Median Rent 1.005 1.005 −0.001 1.485 1.484 −0.000 1.419 1.424 0.005 1.050 1.057 0.007 1.004 1.000 −0.004
Poverty Rate 0.214 0.215 0.001* 0.115 0.116 −0.001 0.198 0.197 −0.001 0.205 0.204 −0.000 0.199 0.200 0.001
% with High School Diploma 0.766 0.758 −0.000 0.902 0.902 0.001 0.753 0.756 0.003 0.818 0.818 −0.001 0.767 0.768 0.000
% Hispanic 0.431 0.450 −0.000 0.123 0.123 −0.001 0.487 0.483 −0.005 0.322 0.319 −0.003 0.487 0.485 −0.002
34 %Non-Hispanic White 0.130 0.121 −0.000 0.519 0.518 0.000 0.264 0.268 0.004 0.256 0.257 0.001 0.207 0.208 0.001
% Black 0.374 0.373 −0.000 0.101 0.102 −0.000 0.090 0.089 −0.001 0.346 0.345 −0.001 0.236 0.239 0.003
% Asian 0.050 0.041 0.000 0.176 0.176 0.000 0.128 0.130 0.002 0.056 0.058 0.003 0.053 0.052 −0.001
Experian Linked: Jan 2020
Credit Score 554 553 −1.798 583 575 8.510* 637 636 −0.961 603 610 6.096** 644 651 4.399
Balance Across All Trades 21.430 23.168 −1.899**19.314 18.346 2.624 19.775 20.011 0.236 32.008 34.483 2.320 19.114 22.391 2.913
Balance in Collections 3.212 3.264 0.007 1.436 1.829 −0.398** 0.917 0.932 0.015 1.121 1.133 0.001 0.491 0.432 −0.046
Balance in Utility Collections 0.576 0.566 0.025 0.298 0.386 −0.084** 0.125 0.120 −0.005 0.306 0.311 0.005 0.147 0.114 −0.029
Balance in Delinquent Accts. 10.225 9.911 0.410 7.776 8.355 −0.567 4.986 4.988 0.003 6.753 6.774 −0.040 2.271 2.136 −0.061
Auto Loan or Lease 0.293 0.299 −0.003 0.308 0.334 −0.019 0.279 0.274 −0.005 0.306 0.338 0.031* 0.214 0.227 0.010
Personal Bankruptcy 0.020 0.019 0.001 0.067 0.087 −0.016 0.050 0.043 −0.007** 0.143 0.151 0.008 0.032 0.030 −0.000
Rev. Line of Credit 0.347 0.343 −0.003 0.529 0.499 0.035 0.656 0.658 0.003 0.569 0.594 0.018 0.553 0.569 0.011
Survey
Response Rates 0.185 0.221 0.035***0.235 0.264 0.016 0.129 0.148 0.020***0.283 0.409 0.126***0.175 0.311 0.136***
N† 4,911 13,460 18,390 6,868 5,087 11,955 13,062 11,315 24,377 73,126 1,537 68,075 4,688 1,765 6,414
Notes: Data come from program applications, the 2015-2019 American Community Survey, and credit records from Experian. The sample includes all program applicants for
all variables except credit attributes, for which the sample is restricted to individuals linked to the balanced panel of Experian records. For each site, we report the average
characteristics for individuals not selected by the lottery (Z = 0) and selected by the lottery (Z = 1). Conditional differences in average characteristics between these two
groups come from regressions that control for site-specific design features. See Section 5. All monetary values expressed in 2020 U.S. dollars divided by 1000. Statistical
significance is denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01. †We report the maximum observation counts. Observation counts for Experian characteristics are lower than
those for the rest of the characteristics.
TABLE II
Assistance Recipients and Targeting
Harris County King County Los Angeles Chicago
DOH TRP
Renters Assisted Renters Assisted Renters Assisted Renters Assisted Assisted
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Demographics:
Age 39.447 . 39.288 39.433 41.659 41.367 40.986 37.969 39.428
Female 0.514 0.669 0.485 0.643 0.510 0.545 0.523 0.601 0.562
Household Size 2.869 2.470 2.252 . 2.856 2.535 2.303 2.886 3.994
White 0.517 0.053 0.580 0.239 0.487 0.250 0.473 0.224 0.033
Black 0.273 0.667 0.112 0.359 0.099 0.114 0.341 0.359 0.100
Hispanic 0.441 0.220 0.120 0.147 0.494 0.494 0.249 0.349 0.837
Asian 0.062 0.020 0.187 0.073 0.127 0.114 0.071 0.068 0.029
ACS Zip Code Data:
35
Median Rent 1.139 1.008 1.598 1.479 1.521 1.443 1.177 1.052 0.999
Poverty Rate 0.163 0.214 0.098 0.111 0.180 0.193 0.192 0.209 0.199
% with High School Diploma 0.786 0.760 0.894 0.895 0.772 0.769 0.830 0.821 0.765
% Hispanic 0.408 0.446 0.111 0.135 0.436 0.462 0.247 0.307 0.492
% Non-Hispanic White 0.310 0.121 0.558 0.508 0.318 0.289 0.350 0.253 0.206
% Black 0.184 0.377 0.072 0.099 0.091 0.083 0.308 0.354 0.234
% Asian 0.077 0.041 0.187 0.176 0.122 0.134 0.072 0.065 0.051
Credit Characteristics:
Credit Score 684 566 725 584 710 649 691 613 658
Balance Across All Trades ($1000s) 19.1 27.9 15.4 21.4 16.7 22.1 19.5 30.7 23.6
Balance in Collections ($1000s) 0.721 3.320 0.219 1.651 0.292 0.918 0.327 1.226 0.419
Balance In Delinquent Accounts 1.661 9.426 0.709 7.428 1.167 4.671 1.126 6.626 1.821
Any Auto Loan or Lease 0.355 0.326 0.261 0.349 0.322 0.279 0.250 0.305 0.214
Any Personal Banckruptcy 0.012 0.019 0.026 0.100 0.031 0.046 0.061 0.140 0.031
Any Open Revolving Line of Credit 0.807 0.384 0.906 0.505 0.870 0.703 0.848 0.599 0.583
N 8,245 2,532 10,794 722 1,607
Notes: This table reports mean characteristics for renters and those who received assistance at each of our four sites. For Chicago those assisted through the DOH program
and the TRP program are reported separately. The first panel reports demographic characteristics. The Assisted include those who received assistance despite not being
selected through the lottery. Renter characteristics are from the 2019 American Community Survey, restricted to renters in Harris County, King County, Los Angeles, and
Chicago. The second panel reports ZIP code characteristics. For renters, these are the weighted average of aggregate ZIP (ZCTA) code characteristics for ZIP codes in the
county/city from the 2015-2019 ACS, weighted by the number of renters in the ZIP code. For those assisted, it is the same, but weighted by the number of applicants. The
third panel reports credit characteristics. For the renter columns, these are from a 10% sample from Equifax restricted to renter in the relevant county/city. For those assisted,
these are for those who were linked to Experian. Both are from January 2020.
TABLE III
First Stage
Received Assistance
Harris County King County Los Angeles Chicago DOH Chicago TRP
(1) (2) (3) (4) (5)
Selected by Lottery 0.590*** 0.372*** 0.207*** 0.431*** 0.910***
(0.005) (0.008) (0.006) (0.013) (0.007)
Control Mean 0.000 0.046 0.347 0.001 0.000
Frac. Selected for Assistance 0.742 0.426 0.464 0.010 0.249
F Statistic 17,096 2,442 1,093 47,977 47,332
N 18,439 12,062 24,377 74,663 6,453
Notes: This table reports OLS estimates of α1 in equation (2). The sample includes all program applicants from Harris
County (column (1)), King County (column (2)), Los Angeles (column (3)), Chicago DOH (column (4)) and Chicago TRP
(column (5)). Regressions include site-specific design controls described in Section 5. The control mean reports the share who
received assistance among individuals not selected by the lottery. “Frac. Selected for Assistance” reports the share of
applicants offered treatment. Robust standard errors are reported in parentheses. Statistical significance is denoted by: * p ≤
0.1; ** p ≤ 0.05; *** p ≤ 0.01.
36
TABLE IV
Survey Outcomes
Harris King Chicago: Chicago:
Combined
County County DOH TRP
(1) (2) (3) (4) (5)
In the Last Month:
Expenditures
All Rent Paid 0.076* 0.134** 0.058* 0.059** 0.082***
(0.040) (0.053) (0.031) (0.028) (0.020)
[0.497] [0.368] [0.649] [0.678] [0.548]
All Bills Paid −0.022 0.026 −0.025 0.023 0.001
(0.032) (0.042) (0.033) (0.033) (0.018)
[0.284] [0.203] [0.358] [0.496] [0.335]
Health
COVID Positive (self-tested) −0.110 −0.064 −0.051*** 0.002 −0.056*
(0.105) (0.074) (0.017) (0.018) (0.033)
[0.182] [0.074] [0.101] [0.095] [0.113]
Feeling Anxious −0.069* 0.014 −0.075** 0.002 −0.032*
(0.036) (0.050) (0.033) (0.028) (0.019)
[0.525] [0.633] [0.549] [0.319] [0.506]
Economic Insecurity
Worried about Eviction −0.082** −0.056 −0.029 −0.022 −0.047***
(0.034) (0.053) (0.030) (0.020) (0.018)
[0.320] [0.443] [0.287] [0.121] [0.293]
Experienced Homelessness −0.031 −0.008 −0.022 −0.005 −0.016
(0.030) (0.037) (0.014) (0.017) (0.013)
[0.241] [0.146] [0.063] [0.079] [0.132]
Stayed in a Homeless Shelter −0.016** −0.003 0.008 −0.000 −0.003
(0.008) (0.013) (0.007) (0.003) (0.004)
[0.015] [0.014] [0.005] [0.001] [0.009]
Was Food Insecure −0.048 0.025 −0.039 −0.016 −0.019
(0.033) (0.049) (0.029) (0.026) (0.018)
[0.335] [0.325] [0.270] [0.256] [0.297]
N 1,671 2,621 14,681 1,352 20,325
Notes: This table reports TSLS estimates of β1 in equation (3), for several outcomes: expenditures, health, and economic
insecurity. Outcomes are derived from the surveys as described in Section 3 and Appendix B. Survey timing and treatment
dates appear in Appendix Table E.2. Columns (1)–(4) report the TSLS estimates for each site, reweighting to adjust for
survey nonresponse as described in Section 3. Column (5) reports the combined averages across all sites. Robust standard
errors are reported in parentheses, and the control group mean is reported in brackets (this corresponds to the mean for the
group not offered assistance, i.e., Z = 0). Statistical significance is denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01.
37
TABLE V
Impacts on Credit Outcomes
Harris County King County Los Angeles Chicago DOH Chicago TRP Combined
(1) (2) (3) (4) (5) (6)
Short Run: 2 months after lottery
Credit Score 2.863** −3.413 0.854 −3.494 0.155 −0.607
(1.235) (4.511) (2.480) (2.755) (1.566) (1.234)
[565] [597] [649] [612] [654] [615]
Balance Across All Trades ($1000s) 0.259 0.113 0.929 1.655 −1.351* 0.321
(0.500) (2.392) (1.086) (1.937) (0.742) (0.677)
[24.825] [21.318] [19.628] [33.758] [20.784] [24.063]
Balance in Collections ($1000s) 0.003 −0.233 −0.157** 0.026 −0.008 −0.074*
(0.058) (0.161) (0.070) (0.067) (0.029) (0.040)
[3.258] [1.371] [0.965] [1.120] [0.489] [1.441]
Balance in Utility Collections ($1000s) 0.013 −0.024 −0.000 −0.030 −0.020** −0.012
(0.019) (0.048) (0.015) (0.031) (0.010) (0.013)
[0.617] [0.255] [0.121] [0.322] [0.150] [0.293]
Balance in Delinquent Accounts ($1000s) −0.230 0.639 0.132 −0.384 −0.146 0.002
(0.218) (0.850) (0.374) (0.559) (0.208) (0.225)
[9.459] [6.315] [4.556] [5.812] [2.152] [5.659]
Any Auto Loan or Lease −0.004 −0.013 0.007 −0.004 0.007 −0.001
(0.007) (0.024) (0.015) (0.016) (0.009) (0.007)
[0.300] [0.311] [0.277] [0.293] [0.212] [0.279]
Any Personal Bankruptcy 0.001 −0.003 −0.007* −0.005 0.000 −0.003
(0.001) (0.006) (0.004) (0.007) (0.002) (0.002)
[0.019] [0.065] [0.047] [0.144] [0.032] [0.061]
Any Open Revolving Line of Credit 0.009 −0.023 0.007 0.026 −0.008 0.002
(0.007) (0.028) (0.012) (0.016) (0.008) (0.007)
[0.372] [0.536] [0.667] [0.576] [0.571] [0.544]
Long Run: 10 months after lottery
Credit Score 0.881 −2.633 −0.932 −6.940* 1.253 −1.674
(1.597) (6.344) (3.713) (2.077) (3.279) (1.693)
[575] [604] [659] [624] [663] [625]
Balance Across All Trades ($1000s) −0.243 −3.166 2.237 −1.329 −0.732 −0.647
(0.830) (4.300) (2.054) (2.623) (1.256) (1.129)
[26.756] [23.879] [20.659] [35.893] [21.809] [25.799]
Balance in Collections ($1000s) 0.007 0.011 0.013 0.065 0.006 0.020
(0.084) (0.250) (0.110) (0.101) (0.043) (0.061)
[3.085] [1.385] [0.969] [1.124] [0.501] [1.413]
Balance in Utility Collections ($1000s) −0.005 −0.016 −0.009 −0.044 −0.011 −0.017
(0.022) (0.059) (0.020) (0.038) (0.014) (0.015)
[0.590] [0.229] [0.109] [0.286] [0.129] [0.269]
Balance in Delinquent Accounts ($1000s) 0.116 −0.687 0.574 0.381 −0.103 0.056
(0.262) (1.019) (0.482) (0.669) (0.222) (0.271)
[8.473] [6.176] [4.275] [5.267] [1.973] [5.233]
Any Auto Loan or Lease 0.001 −0.027 −0.013 0.003 −0.001 −0.007
(0.010) (0.039) (0.022) (0.022) (0.014) (0.011)
[0.314] [0.320] [0.285] [0.328] [0.224] [0.294]
Any Personal Bankruptcy −0.001 0.010 −0.014* −0.024** 0.005* −0.005
(0.002) (0.013) (0.008) (0.010) (0.003) (0.004)
[0.019] [0.059] [0.046] [0.141] [0.030] [0.059]
Any Open Revolving Line of Credit −0.018 −0.019 0.021 0.009 −0.012 −0.005
(0.011) (0.042) (0.019) (0.022) (0.011) (0.011)
[0.435] [0.582] [0.676] [0.603] [0.587] [0.577]
N 14,159 1,889 18,559 45,853 2,577 83,037
Notes: This table reports TSLS estimates of β1 in equation (3), for short-run (2 months) and longer-run (10 months) credit
outcomes. Columns (1)–(4) report the TSLS estimates for each site. Column (5) reports the combined results across all of the
locations. The sample is restricted to lottery applicants linked to Experian credit data. All monetary values are expressed in
thousands of 2020 U.S. dollars. Robust standard errors are reported in parentheses, and the control group mean is reported in
brackets (this corresponds to the mean for the group not offered assistance, i.e. Z = 0). Statistical significance is denoted by:
* p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01.
38
TABLE VI
Impacts on Residential Mobility
Harris County King County Los Angeles Chicago DOH Chicago TRP Combined
(1) (2) (3) (4) (5) (6)
Short Run: 2 months after lottery
Change of Address (Infutor) 0.002 −0.002 0.002 0.000 −0.002 0.000
(0.006) (0.016) (0.009) (0.009) (0.003) (0.004)
[0.011] [0.057] [0.012] [0.009] [0.004] [0.019]
Change of Address (Experian) −0.007 0.040 −0.008 0.001 −0.003 0.005
(0.010) (0.036) (0.017) (0.019) (0.008) (0.009)
[0.102] [0.105] [0.066] [0.080] [0.034] [0.077]
Long Run: 10 months after lottery
Change of Address (Infutor) −0.012 0.012 −0.016 −0.022 −0.018*** −0.011
(0.010) (0.029) (0.022) (0.018) (0.007) (0.008)
[0.044] [0.209] [0.069] [0.049] [0.025] [0.079]
Change of Address (Experian) −0.003 0.046 −0.007 −0.000 0.016 0.010
(0.014) (0.053) (0.027) (0.028) (0.014) (0.014)
[0.250] [0.303] [0.187] [0.202] [0.092] [0.207]
N (Infutor) 6,562 5,570 11,864 26,373 1,616 51,985
N (Experian) 14,159 1,889 18,559 45,853 2,577 83,037
Notes: This table reports TSLS estimates of β1 in equation (3), for residential moves in the short run (2 months) and long
run (10 months). Change of address is measured in two ways: by a direct measure of change of address via Infutor, and by a
change in ZIP code in the Experian linked data. The sample consists of the application sample linked to the respective
outcome data source, and varies by outcome. Columns (1)–(4) report TSLS estimates for each site. Column (5) reports the
combined results across all of the locations. Robust standard errors are reported in parentheses, and the control group mean
is reported in brackets (this corresponds to the mean for the group not offered assistance, i.e. Z = 0). Statistical significance
is denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01.
39
TABLE VII
Impacts on Homelessness
King County Chicago DOH Chicago TRP
(1) (2) (3)
Any Homeless Services 0.020** −0.0021 −0.0035**
(0.0083) (0.0055) (0.0017)
[0.027] [0.0092] [0.0055]
{0.048} {0.0109} {0.0045}
Emergency Shelter 0.0028 −0.0003 −0.0003
(0.0039) (0.0015) (0.0004)
[0.0064] [0.0008] [0.0085]
{0.018} {0.0024} {0.0015}
Street Outreach 0.0067* 0.0016 −0.0007*
(0.0036) (0.0021) (0.0004)
[0.0037] [0.0006] [0.0006]
{0.0091} {0.0021} {0.0014}
Diversion/Prevention 0.0081* −0.0019 −0.0022
(0.0045) (0.005) (0.0015)
[0.0070] [0.0078] [0.0038]
{0.019} {0.007} {0.0017}
Long-Term Subsidies 0.0012 −0.00097*** −0.00065*
(0.0059) (0.0002) (0.0004)
[0.015] [0.00042] [0.0006]
{0.021} {0.0006} {0.0006}
Coordinated Entry 0.0010
(0.0024)
[0.0024]
{0.0078}
N 12,148 74,663 6,453
Notes: This table reports TSLS estimates of β1 in equation (3), for measures of homelessness system use in the 9 months after
the lottery in three lottery samples: King County, Chicago DOH, and Chicago TRP. The first row reports impacts on any
appearance in the homelessness system (“Any Homeless Services”), and the subsequent rows report impacts on specific types
of system use. We assign project types to categories to align with local differences in the nature of services and how the data
were provided. “Emergency Shelter” includes Emergency Shelter, Transitional Housing and Safe Haven in Chicago and
Emergency Shelter and Day Shelter in King County. “Diversion/Prevention” refers to diversion programs in King County and
homelessness prevention projects in Chicago. “Long-term Subsidies” includes Rapid Rehousing and Permanent Supportive
Housing in both sites and also includes Transitional Housing in King County. Information on participation in Coordinated
Entry is only available for King County. Columns (1)–(3) report TSLS estimates for each site. Robust standard errors are
reported in parentheses, and the control group mean of the outcome is reported in brackets (this corresponds to the mean for
the group not offered assistance, i.e. Z = 0). The pre-COVID means from March 2019 to February 2020 for the listed
measures are reported in braces. Statistical significance is denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01.
40
TABLE VIII
Impacts on Eviction Filings
Harris County Chicago: DOH Chicago: TRP
Control Mean Treatment Effect Control Mean Treatment Effect Control Mean Treatment Effect
(1) (2) (3) (4) (5) (6)
Evictions Post Lottery:
Two Months 0.0032 −0.0008 0.0004 −0.0004*** 0.0000 −0.0002
(0.0018) (0.0001) (0.0002)
Six Months 0.0097 −0.0017 0.0012 0.0004 0.0002 −0.0004
(0.0029) (0.0018) (0.0003)
Ten Months 0.0121 −0.0022 0.0019 −0.0004 0.0002 −0.0004
(0.0033) (0.0018) (0.0003)
Evictions Post Moratorium:
Two Months 0.0129 −0.0022 0.0051 −0.0030* 0.0004 0.0000
(0.0034) (0.0022) (0.0007)
Six Months 0.0144 −0.0023 0.0078 −0.0024 0.0004 0.0007
(0.0036) (0.0033) (0.0010)
Ten Months 0.0156 −0.0029 0.0118 −0.0050 0.0011 0.0007
(0.0037) (0.0038) (0.0012)
N 18,390 55,519 6,414
Notes: This table reports TSLS estimates of β1 in equation (3), for measures of eviction filings in three lottery samples:
Harris County, Chicago DOH, and Chicago TRP. The top panel reports impacts on cumulative eviction filings by month
relative to the lottery. The bottom panel summarizes impacts on cumulative eviction filings by month relative to the
expiration of the CDC eviction moratorium. Columns (1), (3), and (5) report the mean among those assigned to the control
group (i.e., those for whom Z = 0). Columns (2),(4), and (6) report TSLS estimates for each site. Robust standard errors are
reported in parentheses, and the control group mean of the outcome is reported in brackets (this corresponds to the mean for
the group not offered assistance, i.e. Z = 0). Statistical significance is denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01.
41
FIGURE I
Timing of ERA Programs Relative to Macroeconomic Trends
Unemployment Rate
A.This data comes from
Unemployment Rate
U.S. Bureau of Labor Statistics.
18 7
Chicago TRP
Households Assisted (Millions)
Chicago LA
15 DOH Harris County 6
Unemployment (%)
King County
5
12
4
9
3
6
2
3
1
0
31 31 31 31 31 31 31 31 31
01− 07− 01− 07− 01− 07− 01− 07− 01−
9− 9− 0− 0− 1− 1− 2− 2− 3−
201 201 202 202 202 202 202 202 202
Unemployment Rate (%) Cumulative Households Assisted, ERA1 & ERA2
Monthly Evictions by Site
Evictions/ Month
Income Relative
($)to January 2020
Eviction data for Chicago, King County, Harris County, and Los Angeles.
Disposable Income (Bottom 50% of Households)
Households
Households
AssistedAssisted
(Millions)
(Millions)
150 B. Household Factor Income and Estimated Disposable Income 7
DOH LA LA
This data comes from realtimeinequality.org.
Harris County
Chicago DOH
125 6
Harris County
4500 7
Chicago
4000 5
Chicago TRP
100 6
Chicago TRP
King County
3500 4
King County
75 5
3000 3
50 4
2500 2
25
2000 1
3
1500
2
1000
31 31 31 31 31 31 31 31 31
01− 07− 01− 07− 01− 07− 01− 07− 01− 1
500
9− 9− 0− 0− 1− 1− 2− 2− 3−
0 201 201 202 202 202 202 202 202 202
31 31 31 31 31 31 31 31 31
01− 07− 01− 07− 01− 07− 01− 07− 01−
9− 9− Chicago0− King County
0− 1− 1− 2− Harris County
2− 3−
201 201 202 202 202 202 202 202 202
Los Angeles Cumulative Households Assisted, ERA1 & ERA2
Factor Income Post−Tax Disposable Income Cumulative Households Assisted, ERA1 & ERA2
Rental Arrears and Eviction Risk
This data is estimated based on the U.S. Census Bureau's Household Pulse Survey.
Households Assisted (Mil
14 7
LA
King County
ouseholds (Millions)
12 6
10 5
Chicago DOH
8
42 4
Chicago TRP
6 3
Harris Co
4 2
3
ted (Millions)
Incom
1500
2
1000
1
500
0
201 201 202 202 202 202 202 202 202
9− 9− FIGURE I 0− 0− 1− 1− 2− 2− 3−
01− 07− 01− 07− 01− 07− 01− 07− 01−
31 31 31
Timing of ERA Programs Relative to Macroeconomic Trends (cont.) 31 31 31 31 31 31
Factor Income Post−Tax Disposable Income Cumulative Households Assisted, ERA1 & ERA2
Rental Arrears
C. Households Behind on Rent, Worried Aboutand Eviction
Next RiskRent, Worried About Eviction
Month’s
This data is estimated based on the U.S. Census Bureau's Household Pulse Survey.
Unemployment Rate
Households
Households
AssistedAssisted
(Millions)
(Millions)
14 7
LA
This data comes from U.S. Bureau of Labor Statistics.
Harris CountyHarris County
Unemployment
Households
(%)(Millions)
12 6
18 7
Chicago TRPChicago TRP
Chicago
LA
10 5
6
15
8 4
DOH Chicago DOH
5
126 3
King County
King County
4 4
2
9
2 1
3
60
2
3 01− 07− 01− 07− 01− 07− 01− 07− 01−
31 31 31 31 31 31 31 31 31 1
9− 9− 0− 0− 1− 1− 2− 2− 3−
0 201 201 202 202 202 202 202 202 202
31 31 31 31 31 31 31 31 31
01− 07− 01− 07− 01− 07− 01− 07− 01−
9− Households in Rental Arrears at Risk of Eviction
9− 0− 0− 1− Households Unconfident in Next Month's Rent Payment
1− 2− 2− 3−
201 201 202 202 202 202 202 202 202
Households in Rental Arrears Cumulative Households Assisted, ERA1 & ERA2
Unemployment Rate (%) Cumulative Households Assisted, ERA1 & ERA2
MonthlyEvictions
D. Monthly Evictions by
by Site
Site
Evictions Relative to January 2020
Eviction data for Chicago, King County, Harris County, and Los Angeles.
Households Assisted (Millions)
150 7
Chicago DOH Chicago TRP
LA
Harris County
125 6
5
King County
100
4
75
3
50
2
25 1
0
31 31 31 31 31 31 31 31 31
01− 07− 01− 07− 01− 07− 01− 07− 01−
9− 9− 0− 0− 1− 1− 2− 2− 3−
201 201 202 202 202 202 202 202 202
Chicago King County Harris County
Los Angeles Cumulative Households Assisted, ERA1 & ERA2
Notes: This figure shows the timing of the ERA programs we study relative to macroeconomic indicators. Panel A
plots the civilian unemployment rate as measured by the Current Population Survey (CPS). Panel B shows Factor
Income and Post-Tax disposable income using estimates from Blanchet et al. (2022). Panel C is based on the authors’
calculations using the Census Household Pulse Surveys. Panel D shows eviction filings for all sites except Los Angeles
where we use completed evictions because filings data are not available. These series are based on administrative
court records. The dashed vertical lines indicate the start of the CARES Act moratorium and the end of the CDC
Eviction Moratorium. ERA1 and ERA2 cumulative households assisted are constructed from the U.S. Treasury
Department’s ERA1 and ERA2 Grantees Compliance Reports.
43
FIGURE II
Targeting: ERA Recipients Vs. Economic Impact Payments Round 1 Recipients
(ZIP-Level Poverty Rate)
Chicago DOH
15%
10%
5%
0%
Chicago TRP
15%
10%
5%
0%
Harris County
15%
Percentage
10%
5%
0%
King County
15%
10%
5%
0%
Los Angeles
15%
10%
5%
0%
0% 10% 20% 30% 40%
Zip Code % Below Poverty Level
EIP r1 recipients ERA recipients
Notes: This figure shows the distribution of ZIP-level population poverty status rates for Emergency Rental Assistance
recipients (cyan) and Economic Impact Payments Round 1 recipients (yellow). The poverty rate measure is from the 2020
American Community Survey (ACS) 5-Year Estimates and is the percentage of the population for whom poverty status is
determined (U.S. Census Bureau, 2020b). EIP r1 data is from the 2020 Internal Revenue Service Statistics of Income and
measures the number of income tax returns filed in 2020 that received round one EIP provided under the CARES Act
(Internal Revenue Service, 2020). The vertical lines indicate the mean poverty rate for each group within each site.
44
FIGURE III
Credit Measure by Month Relative to Application
Credit Score Delinquent Accounts Collections: Utilities
A. Harris County
580 12000 720
680
11000
570
640
10000
600
560
9000 560
550 520
8000
480
540
-10 03/20 -6 -4 -2 0 2 4 6 8 10 12
7000 440
-10 03/20 -6 -4 -2 0 2 4 6 8 10 12 -10 03/20 -6 -4 -2 0 2 4 6 8 10 12
B. King County
610 10000 460
420
9000
600
380
8000
340
590
7000 300
580 260
6000
220
570
-8 03/20 -4 -2 0 2 4 6 8 10 5000 180
-8 03/20 -4 -2 0 2 4 6 8 10 -8 03/20 -4 -2 0 2 4 6 8 10
C. Los Angeles
670 7000 250
210
6000
660
170
5000 130
650
4000 90
640 50
3000
10
630
-6 03/20 -2 0 2 4 6 8 10 12 14 16 2000 -30
-6 03/20 -2 0 2 4 6 8 10 12 14 16 -6 03/20 -2 0 2 4 6 8 10 12 14 16
D. Chicago DOH
640 360
7000
630 320
6000
620
280
5000
610
4000 240
600
3000 200
-2 03/20 2 4 6 8 10 12 14 16 18 20
-2 03/20 2 4 6 8 10 12 14 16 18 20 -2 03/20 2 4 6 8 10 12 14 16 18 20
E. Chicago TRP
680 3000 200
670 160
660 2000 120
650
80
640
-6 03/2020 -2 0 2 4 6 8 10 12 14 16 1000 40
-6 03/20 -2 0 2 4 6 8 10 12 14 16 -6 03/20 -2 0 2 4 6 8 10 12 14 16
Notes: This figure shows the average credit outcome for the group selected for treatment (black diamonds, Z = 1) and the
group not selected for treatment (teal circles, Z = 0), by month relative to the time of application. Month 0 represents the
site’s application month. The dashed vertical line indicates the month of application, and the red vertical line indicates March
2020. The sample includes applicants linked to Experian credit data. Vertical bars plot the 95 percent confidence intervals of
the estimates.
45
FIGURE IV
Reduced-form (ITT) & TSLS Estimates by Month Relative to Application
Credit Score Delinquent Accounts Collections: Utilities
Harris County
20 3000 125
15
2000
10 50
5 1000
0
0 0 -25
-5
-1000
-10 -100
-15 -2000
-20 -3000 -175
0 2 4 6 8 10 12 0 2 4 6 8 10 12 0 2 4 6 8 10 12
King County
20 3000 125
15
2000
10 50
5 1000
0
0 0 -25
-5
-1000
-10 -100
-15 -2000
-20 -3000 -175
0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10
Los Angeles
20 3000 125
15
2000
10 50
5 1000
0
0 0 -25
-5
-1000
-10 -100
-15 -2000
-20 -3000 -175
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Chicago DOH
20 3000 125
15
2000
10 50
5 1000
0
0 0 -25
-5
-1000
-10 -100
-15 -2000
-20 -3000 -175
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Chicago TRP
20 3000 125
15
2000
10 50
5 1000
0
0 0 -25
-5
-1000
-10 -100
-15 -2000
-20 -3000 -175
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16
Notes: This figure shows the reduced-form (ITT) and TSLS estimates for different credit outcomes by month relative to the
application date. The former corresponds to OLS estimates of γ1 in equation (1), while the latter corresponds to TSLS
estimates of β1 in equation (3). The sample includes applicants linked to Experian credit data. Vertical bars plot the 95
percent confidence intervals of the estimates.
46
A Additional Program Details
This section provides information on each program’s timing, eligibility requirements, appli-
cation procedure, and payment amounts. Appendix Table E.10 summarizes these program
details.
A.1 Chicago
We study two assistance lotteries that were run in Chicago in 2020. The first lottery was
run by the City of Chicago Department of Housing (DOH). The second was organized by
The Resurrection Project (TRP), a local nonprofit organization.
Chicago Department of Housing. The City of Chicago Department of Housing (DOH)
partnered with UpTogether to implement an emergency financial assistance program for
city residents at risk of eviction, foreclosure, or homelessness. The program opened to
applications on March 27 and closed on April 5. Applications were submitted through a
simple online form. The program received a total of 75,659 applications, approximately 80
percent of which were from renters, and the remainder from homeowners.
Due to the large number of applicants, grants were distributed through a lottery process.
All applicants were assigned a random rank and were screened for eligibility, and offers were
made in rank order until the funds were exhausted. Upon being selected, applicants had
to provide proof that they were Chicago residents, that the pandemic negatively impacted
their income, and that their pre-pandemic income was no greater than 60 percent of the
area median income (AMI). Ultimately, 722 grants of $1,000 were distributed to eligible
applicants selected by the lottery. Grants were paid directly to tenants on a rolling basis
through April and May of 2020.27
The Resurrection Project. The second program was run by The Resurrection Project
(TRP), a local nonprofit organization, in partnership with the Chicago Resiliency Fund. The
TRP program aimed to provide assistance to a population of people who were not eligible
and did not receive aid from provisions of the CARES Act. Although open to all Chicago
residents with incomes below 300 percent of the federal poverty line, applications from un-
documented individuals, mixed-status families, dependent adults, and returning residents
were particularly encouraged. Those selected had to provide identification, proof of Chicago
residence, and proof of income.
The program’s first round of assistance was not oversubscribed. We study its second
round, which opened for applications on June 22, 2020. Like the DOH program, the TRP
program was open to homeowners and renters, with 75 percent of the applications coming
from renters. In the second round, 10,300 applications were submitted, and 1,607 grants of
$1,000 were distributed directly to tenants by lottery.28
27
DOH conducted two further waves of grant programs later in 2020, but the applicant pools were smaller,
leaving no, or very small, control groups. Therefore, we include only the first round in this study.
28
In total, TRP distributed more than nine million dollars of assistance to over 9,000 people through the
Chicago Resiliency Fund during the pandemic.
47
A.2 Harris County
In November 2020, Harris County, Texas, partnered with Catholic Charities Galveston-
Houston (CCGH) to administer $60 million in assistance to households experiencing eco-
nomic hardship due to the COVID-19 pandemic.29 Eligibility was limited to individuals
earning less than 60 percent of local AMI or receiving public assistance, and who could
demonstrate that the pandemic negatively impacted their income. When applying online,
applicants could upload proof of hardship (e.g., a letter of termination/furlough, a paystub
showing a reduction in pay or hours) or self-attest to hardship using a self-certification form.
Upon selection for processing, applicants underwent an income eligibility screening.
Applications for assistance opened for five days and ended on November 5, 2020. The
county sought to distribute funds equally across four county precincts and therefore capped
the amount of assistance going to each precinct. With this constraint, there was excess
demand in the lowest-income precinct (Precinct 1). A lottery was used to select among
applicants. To prioritize more vulnerable applicants, the lottery featured varying odds of
receiving assistance depending on the applicant’s Census tract.30 Based on the CDC’s Social
Vulnerability Index, the census tracts were divided into quartiles, with the odds of selection
depending on the quartile. Applicants in the top quartile had 50 percent greater odds of
selection. Applicants selected through the lottery and determined eligible received $1,200 in
cash assistance paid directly to the tenant. The funds were mostly distributed in November
and December 2020, with some payments going out in early 2021.
A.3 King County
In late 2020, King County, Washington, implemented an emergency financial assistance pro-
gram through its Department of Community and Human Services (DCHS). DCHS operated
separate programs for large-scale landlords, who could apply directly for assistance for all
eligible tenants, and tenants of small-scale landlords, who could request assistance directly.
We study the latter program. Tenants requested assistance through an online form that was
open from August 20, 2020, to December 4, 2020.
Starting on September 22, 2020, all eligible applicants who had not been offered previous
assistance were entered into a weekly lottery conducted by DCHS. The number of applicants
selected weekly depended on DCHS’s staff capacity. Since people who completed the interest
form earlier were eligible for more weekly lotteries, the probability of treatment varies with
the week of application. The funds for the program were distributed from October 2020 to
January 2021.
To be eligible for this program, individuals had to verify income at or below 50 percent
of the local AMI during the 60 days prior to the application, demonstrate experience of
financial hardship due to COVID-19 that threatened their ability to pay rent on time, and
indicate risk of experiencing housing instability.
The assistance provided in King County was somewhat more generous than that provided
29
This was the second round of assistance distributed by the county, which administered an earlier round
in July 2020.
30
We account for this in our analysis by including Census tract fixed effects (see Section 5.1 for more
detail).
48
at the other sites. Those selected received an average of three months’ back rent or credit
toward future rent, paid to the landlord. The maximum available support covered six months’
rent. To receive this payment, the landlord needed to agree to the following conditions: (i)
to be paid reduced rent equal to the lesser of 80 percent of contracted rent or HUD fair
market rent, (ii) to forgive any rental debt owed by the tenant beyond six months’ rent, (iii)
to retain the tenant (except for cause), and (iv) to not raise rent through March 2022.
A.4 Los Angeles
In July 2020, the Los Angeles City Council approved the creation of the Emergency Rental
Assistance Subsidy (ERAS) program. The program was administered by the City’s Housing
and Community Investment Department (HCIDLA) and allocated $103 million in federal
CARES Act and city funding to provide rent subsidies to low-income tenants.
To be eligible for assistance, renters needed to be city residents who could verify a prepan-
demic income at or below 80 percent of local AMI and provide documentation of a COVID-
related loss of income occurring after March 13, 2020. The application period for the program
spanned five days, from July 13-17, 2020. The city received approximately twice as many
applications from renters as it anticipated being able to serve. To allocate funding, the city
entered applicants into a randomized lottery. Those initially selected by the lottery were no-
tified in late July 2020 and asked to submit documentation to prove their eligibility. Those
not initially selected were placed on a waiting list to be contacted only after attempts were
made to process the applications of all those initially selected.
Applicants who were awarded funding received rent subsidies of $2,000 per household to
be applied to current or future rent payments. Initially, the rent subsidies were paid directly
to landlords, who had to agree to the following three conditions to receive payment: (i)
not to impose interest or late fees on rent owed, (ii) not to evict tenants during the local
declaration of emergency due to COVID-19, and (iii) not to impose a rent increase during
the 12-month period following the award of funding. Limited landlord participation led the
city to approve a change to the policy in November 2020 that allowed the subsidy to be
paid directly to the tenant as a one-time grant in the event that the landlord opted out or
did not respond. The first payment was made in September 2020 and the funds were fully
exhausted at the end of December 2020. Ultimately, 56 percent of all grants were paid to
landlords and 44 percent to tenants.
A.5 Details on Eviction Moratoria
National and local eviction moratoria constituted an important feature of the regulatory
environment for our setting. The CARES Act included a 120-day national moratorium
on eviction filings for non-payment of rent and related fees for properties participating in
federal assistance programs or with federally backed financing.31 This measure was in place
from March 27th to July 24th, 2020. Next, the Centers for Disease Control and Prevention
(CDC) issued a national moratorium on evictions for nonpayment of rent and related fees on
September 4, 2020 (McCarty and Perl, 2021), which lasted until July 31, 2021 (McCarty et
31
This set of properties are estimated to include 28–46% of occupied rental units (Stein and Sutaria, 2020).
49
al., 2021).32 The moratorium allowed evictions for reasons other than non-payment, such as
property damage, criminal activity, health and safety issues, and other contractual violations.
The practical implementation of the moratorium varied by state; some judges saw it as a ban
on filing new cases for non-payment, while others viewed it as a ban on enforcing eviction
orders (Sudeall et al., 2023). Consequently, the impact of the federal moratorium differed
by location.
In addition to the national moratorium, local moratoria were implemented in all our
study sites.
Chicago. Chicago renters were covered by an Illinois state-wide eviction moratorium be-
tween March 20, 2020 and October 3, 2021. The moratorium prohibited eviction filings for
nonpayment of rent, but other types of filings were allowed to continue. Chicago was also
subject to an eviction sealing policy that went into effect on May 17th. The policy sealed
cases filed between March 9th, 2021, and March 31st, 2022, removing the cases from the
public record. These cases remained sealed after March 31st, 2022, while new filings were
part of the public record (ILCS 5/9-122).
Harris County. Harris County renters were covered by a state eviction moratorium from
March 19 to May 19, 2020. Eviction hearings were allowed to resume on May 19, 2020, and
evictions were allowed to begin on May 26, 2020. Although the CDC moratorium covered
renters in Harris County, there was significant local concern that Harris County eviction
court judges were not honoring this order.33 Even though law enforcement could not enforce
an eviction order during the national CDC eviction moratorium and new eviction orders were
automatically “stayed” (delayed) (Benfer et al., 2020), new eviction cases could be filed and
new judgments could be issued. Throughout the duration of the CDC moratorium, Harris
County averaged more than 500 new eviction filings per week, approximately 60 percent of
the pre-COVID volume (Hepburn et al., 2020).
King County. King County renters were covered by both state- and city-level eviction
moratoria. A statewide moratorium was instituted from February 29, 2020, through July
31, 2021. The moratorium statute ended in July, but its protections extended through
October, with landlords’ ability to file evictions reinstated on November 1, 2021. Seattle,
Burien, and Kenmore further extended their city-wide moratoria through January 15, 2022.
Los Angeles. LA city renters were covered by a residential eviction moratorium that was
in effect from March 4, 2020, to June 30, 2022. Starting July 1, 2022, this moratorium
applied only to households with income at or below 80 percent of AMI with demonstrated
COVID-19-related financial hardship.
32
The federal government reinstated a more limited moratorium on August 3, 2021, which the Supreme
Court struck down on August 26, 2021.
33
Local news coverage emphasized uncertainty in how the eviction moratoria were interpreted by local
judges (e.g., Schuetz, 2021).
50
B Construction of Survey Outcomes
This appendix describes how each of the survey outcomes studied in this paper is constructed.
Although similar surveys were administered across sites, they were not identical. Below, we
provide exact definitions and across-site harmonization procedures for each survey outcome
variable we consider (i.e., for each survey variable included in Table IV). Our discussion
references question numbers in the survey instruments, which are provided at https://gi
thub.com/robcollinson/covid_era_surveys/.
All Rent Paid
• For King County, this outcome is based on Q30 and Q31 and is the amount of rent paid
in January 2021 divided by the amount of rent owed in January 2021. The “All Rent
Paid” variable is coded as 1 if this ratio is 1, and 0 otherwise.
• For Harris County, this outcome is based on Q29, for which the response is provided
through a sliding percentage bar indicating the proportion of rent paid in January 2021.
The “All Rent Paid” variable is coded as 1 if the bar was set to 100%, and 0 otherwise.
• For Chicago DOH and Chicago TRP, this outcome is based on a question about whether
the individual paid the full amount of rent due (on time or late). For DOH, the reference
month is May 2020 and is based on Q147. For TRP, the reference month is March 2021
and is based on Q229.
All Bills Paid
• For King and Harris Counties, this outcome is based on Q68, which asks separately
if the respondent paid all of their gas, electric, phone, TV/cable, car payment, car
insurance, health insurance, and student loan bills in the past month. An index is then
created with “yes” coded as 1, “some” as 0.5, and “none” as 0. Bills for which the
respondent answered “did not have bill” are omitted from the index.
• For Chicago DOH and TRP, this outcome is based on Q67 and Q68, which ask “In
the past month, did you or your household pay all of your utility bills such as internet,
phone, gas, electricity?” and “In the past month, did you or your household pay all
of your other bills such as care payments, car insurance, health insurance, or student
loans?” We code the outcome as 1 if the answer is “Yes” to both questions.
COVID Positive (Self-Tested) For this outcome, we use the same two questions across
sites: “Have you suspected that you were infected with COVID-19?” and “Have you tested
positive for COVID-19?” The exact time period to which the question refers varies slightly
across sites. For King and Harris Counties, this question begins with “Since March 1st,”
while for Los Angeles, it begins with “Since May,” and for the Chicago TRP survey, it is
“Since March 13th”. The Chicago DOH survey did not ask these questions; instead, we code
this outcome as 1 if the respondent selects “I got sick” in response to the question “How has
the COVID-19 epidemic impacted you?”
51
Feeling Anxious or Depressed Across all sites, the question “How often have you felt
nervous, anxious, or on edge?” from the PHQ4 is used for this outcome, which is coded as 1 if
the respondent reports “more than half the days” or “nearly every day”. For Harris County,
King County, and Chicago, the question refers to the last week, while for Los Angeles, the
question asks about the last two weeks.
Worried about Eviction For King and Harris Counties, we use the question “How wor-
ried are you about being evicted or foreclosed on in the next three months?” We code the
outcome as 1 if the respondent reports “very worried” and zero otherwise. For all other sites,
we use the same question, but it asks about the next two rather than three months.
Experience Homelessness Across all sites, this outcome is based on a question that
asks whether the respondent has spent any nights couchsurfing, in a homeless shelter, on
the street, in an abandoned building, in a car or van, or in a hotel or motel (for nontravel
reasons). The outcome is coded as “yes” if the respondent answers “yes” to any of the the
above options. For King County, Harris County, Chicago DOH, and Chicago TRP, this
question refers to “the last month”. For Los Angeles, the question is phrased as “Since May,
...”.
Stayed in Shelter This outcome is constructed the same as the “Experience Homeless-
ness” outcome, but it is coded as 1 only if the respondent answers “yes” to the “in a homeless
shelter” option. Similarly to the “Experience Homelessness” outcome, for Los Angeles, the
phrasing refers to “since May”, while for all other sites, the question asks about the last
month.
Was Food Insecure The question “For the past month, which of these statements best
describes the food eaten in your household?” was asked across all locations and coded as 1
if the responded selected “sometimes not enough to eat” or “often not enough to eat”.
52
C Pre-Analysis Plan
The lotteries analyzed in this paper were originally studied by three independent research
teams. The studies were combined and attempts were made to harmonize approaches across
sites as much as possible. Once linked data became available for analysis, pre-registrations
existed for three out of five lotteries. These can be found in the American Economic Asso-
ciation’s RCT Registry: the study IDs are AEARCTR-0007731 (Chicago) and AEARCTR-
0007167 (King County). Of these, the King County registration had a detailed pre-analysis
analysis plan (PAP), which can be downloaded from the registry.
The analysis in this paper follows the PAP for King County. Given the urgency in getting
surveys to respondents once a lottery was implemented by the local partner, it was not always
possible to field the exact same questions across all sites. To facilitate harmonizing survey
outcomes across sites, we deviated in constructing outcomes in the following ways:
• All Rent Paid : While the PAP specified using the fraction of rent paid as a primary
outcome, we use a binary variable indicating whether rent was paid in full. This was
done to create a harmonized measure of rent payment across sites. As described in
Appendix B, three out of five surveys asked about full rent payment (a yes/no question),
while the others asked about the fraction of rent paid. To present a single measure across
the different sites, the fraction of rent was binarized.
• Housing Stability Outcomes: This paper presents estimates for homelessness and ad-
dress changes separately, while the PAP specified as the primary outcome a binary
variable that indicated either of these two outcomes. Because homelessness data was
not available in all sites, it was not possible to construct such an outcome everywhere.
Hence, we opted to present the outcomes separately. Each outcome was listed as a
secondary outcome in the PAP.
• Bills Paid : The PAP designated as our primary outcome the fraction of bills paid.
Similar to the All Rent Paid variable, we instead used a binary outcome indicating
whether all bills were paid, so that we could harmonize across sites. The construction
of this variable is described in Appendix B.
• Feeling Anxious or Depressed : As described in Appendix B, we construct a binary
variable based on multiple choice responses. By contrast, the PAP describes using the
number of days during which the respondent indicated feeling anxious or depressed as
the outcome. Because surveys in different sites asked about different time spans, and
to make the outcome easier to interpret, we harmonized these outcomes across sites by
binarizing them as described in B.
We also harmonized control variables across sites. The King County PAP indicated that, be-
yond design-based controls, we would select additional covariates from the following: lagged
outcomes, gender, race/ethnicity, age, city, disability status, and preferred language. We
have limited these additional controls to those which overlap across sites: gender and race.
53
D Linking Eviction Data
Harris County. We collect eviction case data from 2019 to December 2023 from the
Harris County Justice of the Peace. These data include case information such as case date
and judgment, along with defendant name and address.
We link the lottery applications to eviction records using the following process:
1. Geocode lottery applications and eviction records
2. Construct soundex of first names and last names in lottery applications and eviction
records
3. Exact match on 9-digit ZIP code, soundex of last name and soundex of first name
Approximately 2.65 percent of applicants can be linked to one or more eviction records from
2019 to 2023. Our analysis focuses on the linking to any eviction filing at 2 months, 6
months, and 10 months after the lottery, and 2, 6, and 10 months after the expiration of the
CDC moratorium.
Cook County. Our data is a combination of data from Record Information Services and
data received from the Cook County Clerk’s office, collectively covering the period between
2019 and April 2024, and including case filings that were sealed from the public record
during the eviction moratorium. The data include filing date, each defendant’s name, and
the address of the filing.
We link the records to applications to the DOH and TRP programs as follows:
1. Clean name fields by removing prefix/suffixes and omitting initials.
2. Geocode addresses via ArcGIS Street maps Premium (geocode rates ¿ 97%).
3. Block data by latitude/longitude match up to three decimal places (equal to 1 km)
and conduct fuzzy matching algorithm using Jaro-winkler string distance.
4. Retain only matches with sufficiently similar name field and where geocoordinates are
within 33 meters.
5. Take best match for each eviction filing by name distance and geodistance.
54
E Additional Tables and Figures
55
Appendix Table E.1
Summary and Balance: LA Financial
Los Angeles
Z=0 Z=1 Diff
(1) (2) (3)
Financial Characteristics
Income 29.533 29.835 0.301
Rent 1.614 1.624 0.009
Rent Owed 2.509 2.580 0.071*
N 13,062 11,315 24,377
Notes: Data on financial characteristics come from applications to the LA program. The sample includes all program
applicants. We report the average characteristics for individuals selected by the lottery in column (1) (Z = 1), and the
average characteristics for individuals not selected by the lottery in column (2) (Z = 0). Conditional differences (column (3))
in average characteristics between these two groups come from regressions that control for design features specific to the LA
program. Statistical significance is denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01. We report the maximum observation
counts.
56
Appendix Table E.2
Treatment and Survey Timing
Site Treatment dates Survey dates
King County 09/2020–12/2020 02/2021–03/2021
Harris County 11/2020–01/2021 05/2021–06/2021
Los Angeles 09/2020–12/2020 01/2021–02/2021
Chicago DOH 04/2020–05/2020 05/2020–06/2020
Chicago TRP 08/2020–10/2020∗ 02/2021–04/2021
Notes: This table documents the program and survey timing for each site. For Harris
County, we study the second round of assistance that was offered in November 2020 rather
than the earlier round distributed in July 2020. For Chicago, we study the first round of
the Department of Housing (DOH) program, which provided grants between April and
May 2020, and the second round of The Resurrection Project (TRP) grants program,
which opened to applications in late June 2020.
57
Appendix Table E.3
Survey Outcomes – No Weights
Harris County King County Chicago: DOH Chicago: TRP Combined
(1) (2) (3) (4) (5)
In the Last Month:
Expenditures
All Rent Paid 0.056 0.125** 0.057* 0.051* 0.072***
(0.036) (0.052) (0.030) (0.027) (0.019)
All Bills Paid −0.007 0.023 −0.024 0.024 0.004
(0.029) (0.042) (0.033) (0.032) (0.017)
Health
COVID Positive (self-tested) −0.091 −0.066 −0.051*** −0.003 −0.053**
(0.084) (0.063) (0.016) (0.017) (0.027)
Feeling Anxious −0.049 0.007 −0.072** 0.006 −0.027
(0.032) (0.049) (0.032) (0.027) (0.018)
Economic Insecurity
Worried about Eviction −0.072** −0.061 −0.029 −0.015 −0.044**
(0.030) (0.052) (0.029) (0.020) (0.017)
Experienced Homelessness −0.027 −0.006 −0.022* −0.003 −0.015
(0.026) (0.036) (0.013) (0.016) (0.012)
Stayed in a Homeless Shelter −0.011* −0.004 0.009 0.001 −0.001
(0.007) (0.012) (0.007) (0.003) (0.004)
Was Food Insecure −0.042 0.022 −0.038 −0.013 −0.018
(0.030) (0.048) (0.028) (0.025) (0.017)
N 1,671 2,626 14,681 1,352 20,330
Notes: This table reports the effects of assistance on expenditures, health, and economic insecurity with no reweighting of the
survey responses. Outcomes are derived from the surveys described in Section 3. Survey timing and treatment dates appear in
Appendix Table E.2. See Appendix B for details on the survey questions. Columns (1)–(5) report separately by site the IV
estimates of the effects of assistance on the measure listed in the row, as described in Section 3. Columns (6) and (7) report
the combined averages of all sites and all sites excluding LA, respectively. Robust standard errors are reported in parentheses,
and the control group mean is reported in brackets. Statistical significance is denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤
0.01.
58
Appendix Table E.4
LA Survey Outcomes – Weights and No Weights
Los Angeles
Weights No Weights
(1) (2)
In the Last Month:
Expenditures
All Rent Paid −0.264* −0.215
(0.136) (0.134)
[0.385] [0.385]
All Bills Paid 0.246* 0.239*
(0.140) (0.135)
[0.361] [0.361]
Health
COVID Positive (self-tested) 0.065 0.117
(0.135) (0.129)
[0.263] [0.263]
Feeling Anxious 0.040 0.041
(0.140) (0.136)
[0.543] [0.543]
Economic Insecurity
Worried about Eviction 0.141 0.115
(0.137) (0.130)
[0.343] [0.343]
Experienced Homelessness −0.095 −0.083
(0.079) (0.075)
[0.086] [0.086]
Stayed in a Homeless Shelter 0.000 0.002
(0.016) (0.014)
[0.003] [0.003]
Was Food Insecure −0.004 −0.053
(0.130) (0.122)
[0.291] [0.291]
N 3,360 3,360
Notes: Robust standard errors are reported in parentheses, and the control group mean is reported in brackets. Statistical
significance is denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01.
59
Appendix Table E.5
Balance Table – Survey Response
Harris County King County Los Angeles Chicago DOH Chicago TRP
Z=0 Z=1 Diff Z=0 Z=1 Diff Z=0 Z=1 Diff Z=0 Z=1 Diff Z=0 Z=1 Diff
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15)
Demographics
Age − − − 37.175 37.705 0.530 42.070 41.082 0.988**38.731 39.028 0.297 39.854 39.122 −0.732
60 Household Size − − − − − − 2.605 2.654 −0.049 2.980 2.802 −0.178** 4.043 4.004 −0.039
White 0.071 0.073 −0.008 0.297 0.286 −0.011 0.240 0.237 0.003 0.131 0.132 0.001 0.036 0.022 −0.013
Black 0.616 0.588 0.012 0.268 0.313 0.045** 0.127 0.101 0.026** 0.383 0.333 −0.050** 0.097 0.089 −0.008
Hispanic 0.239 0.272 0.004 0.157 0.151 −0.005 0.519 0.543 −0.024 0.412 0.450 0.038* 0.843 0.857 0.015
Asian 0.014 0.030 0.016** 0.078 0.072 −0.006 0.095 0.089 0.006 0.039 0.060 0.021** 0.025 0.032 0.007
N† 804 536 1,410 1,456 1,170 2,626 1,680 1,680 3,360 14,114 567 14,681 813 539 1,352
Notes: This table reports separately by site the average of demographic characteristics for applicants not selected by the lottery (Z = 0) and applicants selected by the lottery
(Z = 1). Conditional differences in average characteristics (Diff) between these two groups come from regressions that control for site-specific design features. See Section 5.
The sample includes only applicants who completed the survey. The Harris County survey did not ask about age or household size, and the King County survey did not ask
about household size. X Was there another reason why we had the note on observations? X Statistical significance is denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01.
Appendix Figure E.1
Targeting: ERA Recipients Vs. Economic Impact Payments Round 1 Recipients
(ZIP-Level Minority Share)
Chicago DOH
20%
10%
0%
Chicago TRP
20%
10%
0%
Harris County
Percentage
20%
10%
0%
King County
20%
10%
0%
Los Angeles
20%
10%
0%
25% 50% 75% 100%
Zip Code Non−white & Non−Hispanic Population Share
EIP r1 recipients ERA recipients
Notes: This figure shows the distribution of ZIP-level minority share for Emergency Rental Assistance recipients (cyan) and
Economic Impact Payments Round 1 recipients (yellow). The minority share measure is from the 2020 Census Demographic
and Housing Characteristics File and is calculated as the percentage of the population within a ZIP code that identifies as a
minority (U.S. Census Bureau, 2020a). Minority status is determined based on self-reported race and ethnicity, with
individuals identifying as non-white or Hispanic considered to be minorities. EIP r1 data is from the 2020 Internal Revenue
Service Statistics of Income and measures the number of income tax returns filed in 2020 that received round one EIP
provided under the CARES Act (Internal Revenue Service, 2020). The vertical lines indicate the mean minority share for
recipients in each group within each site.
61
Appendix Figure E.2
Targeting: ERA Recipients Vs. Economic Impact Payments Round 1 Recipients
(ZIP-Level Hispanic Share)
Chicago DOH
20%
15%
10%
5%
0%
Chicago TRP
20%
15%
10%
5%
0%
Harris County
20%
Percentage
15%
10%
5%
0%
King County
20%
15%
10%
5%
0%
Los Angeles
20%
15%
10%
5%
0%
0% 25% 50% 75% 100%
Zip Code Hispanic Population Share
EIP r1 recipients ERA recipients
Notes: This figure shows the distribution of ZIP-level Hispanic share for Emergency Rental Assistance recipients (cyan) and
Economic Impact Payments Round 1 recipients (yellow). The Hispanic share measure is from the 2020 Census Demographic
and Housing Characteristics File and is calculated as the percentage of the population within a ZIP code that identifies as
Hispanic (U.S. Census Bureau, 2020a). Hispanic status is determined based on self-reported ethnicity. EIP r1 data is from the
2020 Internal Revenue Service Statistics of Income and measures the number of income tax returns filed in 2020 that received
round one EIP provided under the CARES Act (Internal Revenue Service, 2020). The vertical lines indicate the mean
Hispanic share for recipients in each group within each site.
62
Appendix Figure E.3
Targeting: ERA Recipients Vs. Economic Impact Payments Round 2 Recipients
(ZIP-Level Poverty Rate)
Chicago DOH
15%
10%
5%
0%
Chicago TRP
15%
10%
5%
0%
Harris County
15%
Percentage
10%
5%
0%
King County
15%
10%
5%
0%
Los Angeles
15%
10%
5%
0%
0% 10% 20% 30% 40%
Zip Code % Below Poverty Level
EIP r2 recipients ERA recipients
Notes: This figure shows the distribution of ZIP-level population poverty status rates for Emergency Rental Assistance
recipients (cyan) and Economic Impact Payments Round 2 recipients (yellow). The poverty rate measure is from the 2020
American Community Survey 5-Year Estimates and is the percentage of the population for whom poverty status is
determined (U.S. Census Bureau, 2020b). EIP r2 data is from the 2020 Internal Revenue Service Statistics of Income and
measures the number of income tax returns filed in 2020 that received round two EIP provided under the COVID-related Tax
Relief Act of 2020 (Internal Revenue Service, 2021). The vertical lines indicate the mean poverty rate for recipients in each
group within each site.
63
Appendix Figure E.4
Targeting: ERA Recipients Vs. Economic Impact Payments Round 3 Recipients
(ZIP-Level Poverty Rate)
Chicago DOH
15%
10%
5%
0%
Chicago TRP
15%
10%
5%
0%
Harris County
15%
Percentage
10%
5%
0%
King County
15%
10%
5%
0%
Los Angeles
15%
10%
5%
0%
0% 10% 20% 30% 40%
Zip Code % Below Poverty Level
EIP r3 recipients ERA recipients
Notes: This figure shows the distribution of ZIP-level population poverty status rates for Emergency Rental Assistance
recipients (cyan) and Economic Impact Payments Round 3 recipients (yellow). The poverty rate measure is from the 2020
American Community Survey 5-Year Estimates and is the percentage of the population for whom poverty status is
determined. EIP r3 data is from the 2021 Internal Revenue Service Statistics of Income and measures the number of income
tax returns filed in 2021 that received round three EIP provided under the American Rescue Plan Act of 2021 (Internal
Revenue Service, 2021). The vertical lines indicate the mean poverty rate for recipients in each group within each site.
64
Appendix Figure E.5
Moves in Time Period
Harris County Los Angeles
.1 .08
.08 .06
Moves in Time Period Moves in Time Period
.06 .04
.04 .02
Assigned Control Assigned Control
Assigned Treatment Assigned Treatment
.02 0
03/2020 -6 -4 -2 0 2 4 6 8 10 12 03/2020 -2 0 2 4 6 8 10 12 14 16
Months Relative to Application Months Relative to Application
King County Chicago: DOH
.1
.1
.08 .08
Moves in Time Period Moves in Time Period
.06 .06
.04 .04
Assigned Control Assigned Control
Assigned Treatment Assigned Treatment
.02 .02
-8 03/2020 -4 -2 0 2 4 6 8 10 03/2020 2 4 6 8 10 12 14 16 18 20
Months Relative to Application Months Relative to Application
Chicago: TRP
.06
Moves in Time Period
.04
.02
Assigned Control
Assigned Treatment
0
03/2020 -2 0 2 4 6 8 10 12 14 16
Months Relative to Application
Notes: This figure plots the average number of moves for the treatment and control groups against the number of months
since application. Moves are derived from address changes observed in the Experian data. The dashed vertical line indicates
the month of application (month 0). Applications were submitted at different times across sites. March 2020 is marked on the
horizontal axis.
65
Appendix Figure E.6
Credit Outcomes by Month Relative to Application – Program Applicants vs
Local Renter Population
Credit Score Delinquent Accounts Balance in Collections
A. Harris County
700 11000 3500
9000 3000
650 2500
7000
2000
5000
600 1500
3000
1000
550 1000 500
-10 03/20 -6 -4 -2 0 2 4 6 8 10 12 -10 03/20 -6 -4 -2 0 2 4 6 8 10 12 -10 03/20 -6 -4 -2 0 2 4 6 8 10 12
B. King County
750 10000 2500
700 8000 2000
6000 1500
650
4000 1000
600
2000 500
550 0 0
-8 03/20 -4 -2 0 2 4 6 8 10 -8 03/20 -4 -2 0 2 4 6 8 10 -8 03/20 -4 -2 0 2 4 6 8 10
C. Los Angeles
730 6000
1000
4000
680
500
2000
630 0 0
-6 03/20 -2 0 2 4 6 8 10 12 14 16 -6 03/20 -2 0 2 4 6 8 10 12 14 16 -6 03/20 -2 0 2 4 6 8 10 12 14 16
D. Chicago DOH
690 6800 1200
4800
640 700
2800
590 800 200
-2 03/20 2 4 6 8 10 12 14 16 18 20 -2 03/20 2 4 6 8 10 12 14 16 18 20 -2 03/20 2 4 6 8 10 12 14 16 18 20
730
E. Chicago TRP
700
2800
680
630 800 200
-6 03/20 -2 0 2 4 6 8 10 12 14 16 -6 03/20 -2 0 2 4 6 8 10 12 14 16 -6 03/20 -2 0 2 4 6 8 10 12 14 16
Notes: This figure plots the average credit characteristics for the treatment and control groups and for renters against the
number of months since application. Data on local renters come from a 10 percent sample of individuals from Equifax and
include individuals outside of the applicant pool. Control and treatment samples include only applicants linked to the
Experian credit data. Applications were submitted at different times across sites. Each column corresponds to a different
credit characteristic: credit score, balance in delinquent accounts, and balance in utility collections. Each row corresponds to
a different location: Harris County, King County, Los Angeles, Chicago DOH, and Chicago TRP. The dashed vertical line
indicates the month of application, and the red vertical line indicates March 2020.
66
Appendix Figure E.7
Emergency Shelter Entry Rate by Random Assignment and Take-up of Treatment
Notes: This figure plots the rate at which assistance applicants enroll in emergency shelters as tracked by HMIS data. The
sample includes all lottery applicants in King County. Treatment and control groups are split by random assignment; the
treatment group is further split by whether the assistance was successfully paid.
67
Appendix Table E.6
Average Credit Characteristics – Program Applicants vs Local Population
Harris County King County Los Angeles Chicago
DOH TRP
All Renters Assisted All Renters Assisted All Renters Assisted All Renters Assisted Assisted
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)
Credit Characteristics:
Credit Score 707 684 566 748 725 584 727 710 649 712 691 613 658
(94) (94) (78) (75) (80) (85) (84) (85) (1) (95) (97) (93) (89)
Balance Across All Trades ($1000s) 68.8 19.1 27.9 118.8 15.4 21.4 103.2 16.7 22.1 71.9 19.5 30.7 23.6
(155.2) (30.6) (47.4) (263.1) (32.2) (60.8) (345.8) (37.6) (0.7) (161.3) (36.4) (60.7) (54.3)
Balance in Collections ($1000s) 0.591 0.721 3.320 0.168 0.219 1.651 0.229 0.292 0.918 0.281 0.327 1.226 0.419
(2.637) (2.285) (4.299) (1.220) (1.166) (3.451) (2.033) (2.356) (0.038) (1.794) (1.575) (2.346) (1.311)
68
Balance In Delinquent Accounts ($1000s) 1.514 1.661 9.426 0.549 0.709 7.428 1.025 1.167 4.671 1.126 1.126 6.626 1.821
(6.898) (6.966) (12.816) (5.776) (6.549) (20.045) (8.040) (8.407) (0.168) (8.397) (7.045) (15.320) (6.099)
Any Auto Loan or Lease 0.398 0.355 0.326 0.282 0.261 0.349 0.346 0.322 0.279 0.274 0.250 0.305 0.214
(0.490) (0.479) (0.459) (0.450) (0.439) (0.477) (0.476) (0.467) (0.007) (0.446) (0.433) (0.461) (0.410)
Any Personal Banckruptcy 0.012 0.012 0.019 0.019 0.026 0.100 0.026 0.031 0.046 0.050 0.061 0.140 0.031
(0.108) (0.108) (0.137) (0.135) (0.158) (0.300) (0.161) (0.173) (0.003) (0.217) (0.239) (0.348) (0.174)
Any Open Revolving Line of Credit 0.847 0.807 0.384 0.929 0.906 0.505 0.887 0.870 0.703 0.874 0.848 0.599 0.583
(0.360) (0.395) (0.469) (0.256) (0.291) (0.500) (0.316) (0.336) (0.007) (0.332) (0.359) (0.491) (0.493)
Observations: 26,086 14,803 6,169 14,408 7,358 491 24,579 16,603 8,702 14,399 9,200 357 640
Notes: This table reports the average credit characteristics for program applicants and a sample of local residents from outside the applicant pool for each site. Data on local
residents come from a 10 percent sample of individuals from Equifax. “All” includes both renters and homeowners, and “Renters” is limited to individuals who rent.
“Assisted” are the treated individuals who are linked to the Experian credit data at the time = 0. All monetary values are expressed in 2020 U.S. dollars divided by 1000.
Standard deviations are reported in parentheses.
Appendix Table E.7
Treatment Effects on Homelessness System Use (King County and Chicago)
King County King County Chicago DOH Chicago TRP
Full Sample Survey Respondents Full Sample Full Sample
Any Homelessness 0.020** 0.027* -0.00213 -0.00346**
Services (0.0083) (0.016) (0.00546) (0.00172)
[0.027] [0.027] [0.0092] [0.00555]
{0.048} {0.047} {0.0109} {0.00449}
Emergency 0.0028 -0.000065 -0.00028 -0.00026
Shelter (0.0039) (0.0077) (0.00153) (0.00041)
[0.0064] [0.0081] [0.000779] [0.00853]
{0.018} {0.022} {0.00242} {0.00155}
Street 0.0067* 0.015** 0.00157 -0.0007*
Outreach (0.0036) (0.0068) (0.00215) (0.00041)
[0.0037] [0.0023] [0.000602] [0.00064]
{0.0091} {0.0098} {0.00208} {0.00139}
Diversion 0.0081* 0.0020 -0.0019 -0.00225
(0.0045) (0.0095) (0.00502) (0.00148)
[0.0070] [0.010] [0.00782] [0.00384]
{0.019} {0.017} {0.00702} {0.0017}
Any Longer- 0.0012 0.010 -0.00097*** -0.00065*
Term Subsidies (0.0059) (0.011) (0.00018) (0.00039)
[0.015] [0.013] [0.000424] [0.00064]
{0.021} {0.019} {0.000656} {0.00062}
Coordinated 0.00100 0.0013
Entry (0.0024) (0.0038)
[0.0024] [0.0023]
{0.0078} {0.0070}
N 12148 3152 74663 6453
Notes: This table reports the effects of assistance receipt on homelessness services use among King County and Chicago
program applicants. Outcomes were measured 9 months after application based on administrative homelessness management
information system (HMIS) data. Columns (1) and (2) report the IV estimates of the effects of assistance on the measure
listed in the row for the full King County sample and the sample of King County survey respondents, respectively. Columns
(3) and (4) report the IV estimates of the effects of assistance for the Chicago DOH and TRP samples, respectively. Robust
standard errors are reported in parentheses, and the control group means are reported in brackets. The pre-COVID means for
March 2019 to February 2020 are reported in braces. Statistical significance is denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤
0.01.
69
Appendix Table E.8
Additional Survey Characteristics
Harris King Los Chicago Chicago
County County Angeles DOH TRP
(1) (2) (3) (4) (5)
Financial Characteristics:
Monthly Rent 1.021 1.472 1.571 1.122 1.026
Months of Rent Owed 1.090 3.259 1.745 – 0.839
Monthly Income (Feb 2020) 1.498 1.722 – 1.374 1.150
Any Paid Work (Feb 2020) 0.891 0.837 – 0.762 0.759
Demographics:
Age – 37.287 41.576 38.409 39.289
Female 0.783 0.620 0.564 0.637 0.613
Household Size 3.520 2.867 2.629 3.314 3.988
White 0.083 0.343 0.239 0.120 0.021
Black 0.613 0.364 0.114 0.373 0.117
Hispanic 0.271 0.186 0.531 0.430 0.807
N 1422 2621 24374 14681 1352
Notes: This table reports the mean of each characteristic for all who filled out the survey in each site (includes treatment and
control). We report reweighted means to adjust for survey nonresponse. All monetary values (“Monthly Rent” and “Monthly
Income (Feb 2020)”) are divided by 1,000. We report the maximum sample size for each site. Monthly income is household
income in February 2020. Note that, for King County, the variable “Months of Rent Owed” comes from administrative
payment data, not the survey.
70
Appendix Table E.9
Fraction of Participants Receiving Assistance in Future Rounds
King County Chicago: DOH
Assigned: Assigned:
Control Treatment Control Treatment
(1) (2) (3) (4)
Received Assistance in Future Round 0.28 0.31 0.07 0.13
N 6,982 5,166 73,126 1,537
Notes: This table reports the fraction of lottery participants who received assistance in a future round. The definition of the
treatment and control groups is from the round one assignments. For Chicago DOH, these numbers include those who
received assistance from round 2 of DOH or from TRP.
71
Appendix Table E.10
Comparison of Lotteries to Other Emergency Relief Assistance Programs by Site
Chicago DOH Chicago TRP Houston LA Seattle CARES Act ERAP Treasury ERAP†
(1) (2) (3) (4) (5) (6) (7)
Monthly Amount of Assistance $1,000 $1,000 $1,200 $1,000-$2,000 min(80% of rent Median: $1,200 Max: $17,000
or HUD FMR) Mean: $1,700
Range: $300 –
$3,300 (94%)
Duration of Assistance 1 Month 1 Month 1 Month 1 Month 1-6 Months ≤3 Months (60%) Max: 15 Months
(Median:3) 6–12 Months
(30%)
Recipient Tenant Tenant Tenant Tenant or Landlord Landlord (94%) Tenant (71%)
Landlord Tenant (6%)
Application Dates Apr-20 Jul-21 Nov-20 Jul-20 Sept–Nov-20 Mar/Apr-20 (15%) Jan/Mar-21
May/Jun-20 (39%) (ERA1)
Jul/Aug-20 (41%)‡ Mar-21/Jun-22
(ERA2)*
Dates of Assistance Apr–Jun-20 Jul–Aug-20 Nov–Dec-20 Aug–Dec-20 Oct–Dec-20 First-come, first- First-come, first-
served (45%) served (56%)
Lottery (16%) Lottery (3%)
Notes: The data for CARES Act ERAP come from Reina et al. (2021) and Yae et al. (2023) from the National Low Income
Housing Coalition and the Housing Initiative at Penn. The sample of programs in this study includes 220 programs spanning
40 states (including Washington, DC). The percentages in parentheses represent the number of sites in the sample with the
listed characteristics.
† The sample for Treasury ERAP consists of 389 programs across the US. For Monthly Amount of Assistance, we report the
median of the maximum amount of assistance programs reported giving. Data for average monthly assistance across programs
are currently unavailable. For Duration of Assistance, we also report the median of the maximum number of months
programs reported providing assistance.
‡ These dates show the months when the programs first began accepting applications. These percentages are based on 179 of
the 220 total programs. Only 1% of programs began in January of 2020, and 4% in September of 2020.
* The December 2020 Consolidated Appropriations Act of 2021 created the Treasury ERA Program, establishing ERA1. In
March of 2021, Congress provided additional funds to the Treasury ERA Program through the American Rescue Plan Act,
establishing ERA2. June 2022 was the last month for which states with ERA programs were required to provide monthly
reporting to the treasury. ERA spending rapidly increased in the months leading up to the end of the federal eviction
moratorium in August of 2021.
72
Appendix Table E.11
Rental Agreements with Landlords
Harris County King County Chicago: DOH Chicago: TRP
(1) (2) (3) (4)
T
P (Missed Payment Had an Agreement) 0.304 0.487 0.316 0.286
P (Had an Agreement|Missed Payment) 0.721 0.724 0.735 0.880
Agreement with:
Written agreement 0.273 0.277 0.223 0.133
Some rent forgiven 0.073 0.035 0.061 0.040
Had more time 0.741 0.331 0.556 0.715
Had payment plan 0.306 0.161 0.174 0.195
Had more time- late fee 0.306 0.045 0.116 0.074
Late fees waived 0.177 0.146 0.221 0.118
Other 0.086 0.061 0.069 0.059
N 1,675 2,626 14,681 1,352
Notes: This table describes the types of agreements that tenants who missed rent payments reached with their landlords.
These numbers are taken from the answers to Q38–Q40 in the Harris County and King County surveys. The first row is the
fraction of respondents in each site who missed a payment and had an agreement with their landlord. The second row is the
fraction among those respondents who missed a payment who had an agreement with their landlord. The rows that follow
describe the content of these agreements. The categories listed are not mutually exclusive: respondents were instructed to
select all features included in their agreement. These numbers should be interpreted as the fraction of respondents whose
agreement included at least this feature.
73
Appendix Table E.12
Rates of Government Assistance
King County Chicago: DOH Chicago: TRP
(1) (2) (3)
Pre-Pandemic
UI 0.117 0.041 0.014
Disability 0.030 0.018 0.013
Medicaid 0.168 0.196 0.137
Medicare 0.090 0.074 0.070
SNAP 0.293 0.348 0.339
WIC 0.091 0.062 0.057
TANF 0.059 0.013 0.005
Social Security 0.091 0.059 0.050
Community 0.035 – –
Food Bank 0.166 0.048 0.078
Housing – 0.043 0.016
Union – 0.009 0.015
Post-Lottery
UI 0.166 0.140 0.022
Disability 0.013 0.017 0.014
Medicaid 0.101 0.210 0.137
Medicare 0.047 0.080 0.078
SNAP 0.188 0.411 0.367
WIC 0.039 0.062 0.058
TANF 0.032 0.014 0.006
Social Security 0.045 0.059 0.054
Community 0.029 – –
Food Bank 0.086 0.059 0.086
Housing 0.056 0.042 0.017
Union – 0.015 0.015
N 2,626 14,681 1,352
Notes: This table reports the pooled estimates for the treatment and control groups of the rates of government assistance
pre-pandemic and post-lottery for the different sites (based on survey responses). For King County, the data from the
pre-pandemic period are from February 2020 and from August 2020 for the post-lottery period. For Chicago DOH, the
pre-pandemic period is February 2020, and the post-lottery period is April 2020. For Chicago TRP, the pre-pandemic period
is February 2020 and the post-lottery period is February 2021.
74
Appendix Figure E.8
Unemployment Rate by Tenure
0.15
0.10
Percentage
0.05
0.00
2000 2005 2010 2015 2020
Year
Owners Renters
Notes: This figure shows unemployment rates for renters and homeowners. Source: Current Population Survey Data Annual
Social and Economic Supplements (CPS ASEC).
75
F Heterogeneity-Analysis
F.1 Heterogeneity on individual observable characteristics
76
Appendix Figure F.1
Heterogeneity in Financial Health Outcomes by Baseline Financial
Characteristics
Los Angeles King County
Harris County
ALL ALL
ALL Credit Score Credit Score
Credit Score Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Delinquent Accts Balance in Utility Collections Balance in Utility Collections
Balance in Utility Collections
<580 CREDIT SCORE <580 CREDIT SCORE
<580 CREDIT SCORE
Credit Score Credit Score
Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
Balance in Utility Collections
>580 CREDIT SCORE >580 CREDIT SCORE
>580 CREDIT SCORE
Credit Score Credit Score
Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
Balance in Utility Collections
CREDIT SCORE INCREASE CREDIT SCORE INCREASE
CREDIT SCORE INCREASE
Credit Score Credit Score
Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
Balance in Utility Collections
CREDIT SCORE DECREASE CREDIT SCORE DECREASE
CREDIT SCORE DECREASE
Credit Score Credit Score
Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
Balance in Utility Collections
<5K DELINQUENCY <5K DELINQUENCY
<5K DELINQUENCY
Credit Score Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections Balance in Utility Collections
>5K DELINQUENCY >5K DELINQUENCY >5K DELINQUENCY
Credit Score Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections Balance in Utility Collections
DELINQUENCY INCREASE DELINQUENCY INCREASE DELINQUENCY INCREASE
Credit Score Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections Balance in Utility Collections
DELINQUENCY DECREASE DELINQUENCY DECREASE DELINQUENCY DECREASE
Credit Score Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections Balance in Utility Collections
-0.5 0 0.5 -0.5 0 0.5 -0.5 0 0.5
Chicago: DOH Chicago: TRP
ALL ALL
Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
<580 CREDIT SCORE <580 CREDIT SCORE
Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
>580 CREDIT SCORE >580 CREDIT SCORE
Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
CREDIT SCORE INCREASE CREDIT SCORE INCREASE
Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
CREDIT SCORE DECREASE CREDIT SCORE DECREASE
Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
<5K DELINQUENCY <5K DELINQUENCY
Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
>5K DELINQUENCY >5K DELINQUENCY
Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
DELINQUENCY INCREASE DELINQUENCY INCREASE
Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
DELINQUENCY DECREASE DELINQUENCY DECREASE
Credit Score Credit Score
Balance in Delinquent Accts Balance in Delinquent Accts
Balance in Utility Collections Balance in Utility Collections
-0.5 0 0.5 -0.5 0 0.5
Notes: This figure plots the IV estimates of the effects on credit outcomes for different subgroups. The subgroups are in bold
and followed by the three credit outcomes: credit score, balance in delinquent accounts, balance in utility collections. Point
estimates are standardized by dividing the IV estimate by the standard deviation of the outcome in the control group. The
leftmost forest plot is for Harris County and is followed by the plots of Los Angeles and King County. The sample includes
applicants linked to the Experian credit data.
77
Appendix Table F.1
Heterogeneity in Impacts on Expenditures and Health
Expenditures Health
All Rent All Bills COVID Feeling
Paid Paid Positive Anxious
(1) (2) (3) (4)
Mental Health: Good 0.072*** 0.003 −0.031** −0.036*
(0.020) (0.022) (0.014) (0.021)
Mental Health: Bad 0.053 0.005 −0.010 −0.029
(0.034) (0.030) (0.025) (0.033)
Higher Income 0.058** −0.019 −0.028** −0.035
(0.023) (0.024) (0.014) (0.023)
Lower Income 0.060** 0.014 −0.025 −0.022
(0.027) (0.026) (0.022) (0.027)
Some Safety Net 0.082*** −0.016 −0.042*** −0.034
(0.023) (0.023) (0.016) (0.023)
No Safety Net 0.049* 0.023 −0.014 −0.024
(0.026) (0.028) (0.019) (0.027)
Employed 0.026 −0.010 −0.048*** −0.010
(0.027) (0.027) (0.019) (0.026)
Not Employed 0.077*** −0.005 −0.011 −0.059**
(0.025) (0.025) (0.017) (0.026)
Not Behind on Rent 0.071*** 0.002 −0.032** −0.054**
(0.021) (0.022) (0.014) (0.022)
Behind on Rent 0.040 0.018 −0.009 −0.005
(0.030) (0.029) (0.024) (0.030)
Moved Last Yr 0.072*** −0.020 −0.030** −0.026
(0.019) (0.020) (0.013) (0.020)
Not Moved Last Yr 0.048 0.072* −0.013 −0.075*
(0.040) (0.041) (0.033) (0.040)
N 8,594 8,778 8,064 9,554
Notes: This table reports heterogeneity in the effects of assistance on survey measures of expenditures and health outcomes
by reported pre-lottery characteristics. Each column is a different survey outcome. The reported estimates are combined
inverse-variance weighted averages across all non-LA sites. Robust standard errors are reported in parentheses, and the
control group mean is reported in brackets (this corresponds to the mean for the group not offered assistance, i.e., Z = 0).
Statistical significance is denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01.
78
Appendix Table F.2
Heterogeneity in Impacts on Economic Insecurity
Economic Insecurity
Worried Expeienced Stayed in
Was Food
Abt. Homeless- Homeless
Insecure
Eviction ness Shelter
(1) (2) (3) (4)
Mental Health: Good −0.022 −0.019** −0.002 −0.033*
(0.016) (0.009) (0.004) (0.018)
Mental Health: Bad −0.069** −0.012 −0.002 0.002
(0.031) (0.025) (0.007) (0.033)
Higher Income −0.013 −0.017 −0.002 −0.033*
(0.018) (0.011) (0.003) (0.020)
Lower Income −0.077*** −0.017 0.002 0.011
(0.025) (0.019) (0.005) (0.027)
Some Safety Net −0.019 0.002 0.002 −0.005
(0.020) (0.015) (0.003) (0.021)
No Safety Net −0.047** −0.039*** −0.005*** −0.048*
(0.022) (0.011) (0.001) (0.025)
Employed −0.015 −0.024* −0.003 −0.032
(0.023) (0.013) (0.005) (0.024)
Not Employed −0.036* −0.013 0.001 0.011
(0.021) (0.016) (0.003) (0.023)
Not Behind on Rent −0.023 −0.020** 0.003 −0.016
(0.018) (0.009) (0.004) (0.020)
Behind on Rent −0.067*** −0.005 −0.007 −0.038
(0.026) (0.022) (0.006) (0.029)
Moved Last Yr −0.036** −0.024*** −0.003 −0.032*
(0.016) (0.008) (0.004) (0.018)
Not Moved Last Yr −0.051 0.029 0.010 0.020
(0.037) (0.032) (0.011) (0.039)
N 8,478 9,539 9,310 9,562
Notes: This table reports heterogeneity in the effects of assistance on survey measures of economic insecurity by reported
pre-lottery characteristics. Each column is a different survey outcome. The reported estimates are combined inverse-variance
weighted averages across all non-LA sites. Robust standard errors are reported in parentheses, and the control group mean is
reported in brackets (this corresponds to the mean for the group not offered assistance, i.e., Z = 0). Statistical significance is
denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01.
79
Appendix Figure F.2
Survey Heterogeneity: Expenditure and Health Outcomes
Effects on Rent Paid Effects on Bills Paid
-.15 -.1 -.05 0 .05 .1 .15 -.15 -.1 -.05 0 .05 .1 .15
High-Inc High-Inc
Low-Inc Low-Inc
No Safety-Net No Safety-Net
Some Safety-Net Some Safety-Net
Mental Health: good Mental Health: good
Mental Health: poor Mental Health: poor
Employed Employed
Unemployed Unemployed
Behind Rent Behind Rent
Current Rent Current Rent
Moved Last Yr Moved Last Yr
Not Moved Last Yr Not Moved Last Yr
Effects on Covid Test Pos. Effects on Feeling Anxious
-.1 -.05 0 .05 .1
High-Inc
-.15 -.1 -.05 0 .05 .1 .15
Low-Inc High-Inc
Low-Inc
No Safety-Net
No Safety-Net
Some Safety-Net
Some Safety-Net
Mental Health: good
Mental Health: good
Mental Health: poor
Mental Health: poor
Employed
Employed
Unemployed
Unemployed
Behind Rent
Behind Rent
Current Rent Current Rent
Moved Last Yr Moved Last Yr
Not Moved Last Yr Not Moved Last Yr
Notes: This figure plots the combined IV estimates of the effects of assistance across sites on survey outcomes for various
subgroups. The subgroups appear in the rows of each figure. The outcome is listed in the column. The combined estimates
are inverse-variance weighted averages across sites. The whiskers are 90 percent confidence intervals.
80
Appendix Figure F.3
Survey Heterogeneity: Economic Insecurity Outcomes
Effect on Any Homelessness Effects on Food Insecurity
-.1 -.05 0 .05 .1 -.1 -.05 0 .05 .1
High-Inc High-Inc
Low-Inc Low-Inc
No Safety-Net No Safety-Net
Some Safety-Net Some Safety-Net
Mental Health: good Mental Health: good
Mental Health: poor Mental Health: poor
Employed Employed
Unemployed Unemployed
Behind Rent Behind Rent
Current Rent Current Rent
Moved Last Yr Moved Last Yr
Not Moved Last Yr Not Moved Last Yr
Effect on Shelter Use Effects on Worried About Eviction
-.03 -.02 -.01 0 .01 .02 .03 -.15 -.1 -.05 0 .05 .1 .15
High-Inc High-Inc
Low-Inc Low-Inc
No Safety-Net No Safety-Net
Some Safety-Net Some Safety-Net
Mental Health: good Mental Health: good
Mental Health: poor Mental Health: poor
Employed Employed
Unemployed Unemployed
Behind Rent Behind Rent
Current Rent Current Rent
Moved LastYr Moved LastYr
Not Moved LastYr Not Moved LastYr
Notes: This figure plots the combined IV estimates of the effects of assistance across sites on survey outcomes for various
subgroups. The subgroups appear in the rows of each figure. The outcome is listed in the column. The combined estimates
are inverse-variance weighted averages across sites. The whiskers are 90 percent confidence intervals.
81
Appendix Table F.3
Heterogeneity in Impacts on Expenditures and Health (Equally-Weighted by
Lottery)
Expenditures Health
All Rent All Bills COVID Feeling
Paid Paid Positive Anxious
(1) (2) (3) (4)
Mental Health: Good 0.088*** 0.010 −0.033 −0.042*
(0.025) (0.024) (0.024) (0.025)
Mental Health: Bad 0.041 −0.001 −0.017 −0.026
(0.036) (0.031) (0.034) (0.034)
Higher Income 0.077*** −0.012 −0.043 −0.053*
(0.029) (0.027) (0.035) (0.028)
Lower Income 0.063** 0.020 −0.018 −0.010
(0.029) (0.028) (0.026) (0.029)
Some Safety Net 0.090*** −0.013 −0.036 −0.032
(0.026) (0.024) (0.025) (0.026)
No Safety Net 0.067** 0.020 −0.020 −0.025
(0.031) (0.030) (0.028) (0.032)
Employed 0.063** −0.003 −0.053* −0.028
(0.031) (0.028) (0.031) (0.030)
Not Employed 0.077*** −0.004 −0.013 −0.051*
(0.026) (0.025) (0.023) (0.027)
Not Behind on Rent 0.076*** 0.004 −0.006 −0.050*
(0.026) (0.023) (0.027) (0.026)
Behind on Rent 0.050 0.010 −0.026 −0.014
(0.032) (0.029) (0.029) (0.031)
Moved Last Yr 0.085*** −0.017 −0.015 −0.033
(0.023) (0.021) (0.023) (0.023)
Not Moved Last Yr 0.070* 0.081** −0.035 −0.079*
(0.042) (0.041) (0.040) (0.041)
N 8,594 8,778 8,064 9,554
Notes: This table reports heterogeneity in the effects of assistance on survey measures of expenditures and health outcomes
by reported pre-lottery characteristics. Each column is a different survey outcome. The reported estimates are combined
equally-weighted averages across all non-LA lotteries. Robust standard errors are reported in parentheses, and the control
group mean is reported in brackets (this corresponds to the mean for the group not offered assistance, i.e., Z = 0). Statistical
significance is denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01.
82
Appendix Table F.4
Heterogeneity in Impacts on Economic Insecurity (Equally-Weighted by Lottery)
Economic Insecurity
Worried Expeienced Stayed in
Was Food
Abt. Homeless- Homeless
Insecure
Eviction ness Shelter
(1) (2) (3) (4)
Mental Health: Good −0.032 −0.021 −0.006 −0.032
(0.023) (0.017) (0.004) (0.023)
Mental Health: Bad −0.070** −0.015 0.003 0.020
(0.035) (0.026) (0.009) (0.035)
Higher Income −0.019 −0.012 −0.007 −0.047*
(0.026) (0.019) (0.005) (0.025)
Lower Income −0.086*** −0.030 0.000 0.030
(0.028) (0.021) (0.007) (0.028)
Some Safety Net −0.042* −0.007 0.002 0.007
(0.025) (0.019) (0.006) (0.024)
No Safety Net −0.050 −0.038 −0.009 −0.059**
(0.031) (0.024) (0.006) (0.030)
Employed −0.047 −0.011 −0.003 −0.031
(0.029) (0.020) (0.005) (0.028)
Not Employed −0.049* −0.022 −0.005 0.014
(0.025) (0.019) (0.006) (0.025)
Not Behind on Rent −0.051** −0.010 0.007 0.001
(0.025) (0.019) (0.006) (0.024)
Behind on Rent −0.082*** −0.005 −0.004 −0.040
(0.030) (0.024) (0.008) (0.030)
Moved Last Yr −0.075*** −0.022 −0.006 −0.027
(0.022) (0.014) (0.004) (0.021)
Not Moved Last Yr −0.040 0.031 0.013 0.015
(0.040) (0.035) (0.012) (0.040)
N 8,478 9,539 9,310 9,562
Notes: This table reports heterogeneity in the effects of assistance on survey measures of economic insecurity by reported
pre-lottery characteristics. Each column is a different survey outcome. The reported estimates are combined equally-weighted
averages across all non-LA lotteries. Robust standard errors are reported in parentheses, and the control group mean is
reported in brackets (this corresponds to the mean for the group not offered assistance, i.e., Z = 0). Statistical significance is
denoted by: * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01.
83
F.2 Cross-Site Heterogeneity
F.2.1 Methodology and Notation
This appendix section draws on tools from meta-analysis to test for site-specific heterogeneity
using a random-effects framework. Following DerSimonian and Laird (1986) (DL), we model
site-specific heterogeneity as a random effect, which we estimate with noise due to sampling
variation.34 One advantage of this approach is that, as it is a method-of-moments estimator,
we do not need to make distributional assumptions about the true underlying site-specific
effects.
Consider our site-level estimates {β̂i }ki=1 with sampling variances {s2i }. Following a stan-
dard random-effects model, we will decompose each observed estimate into a site-specific
“true” effect plus sampling error:
approx
β̂i = βi + εi , εi ∼ N (0, s2i ),
βi = β + e i , E[ei ] = 0, Var(ei ) = τ 2 ,
approx
so β̂i ∼ N (β, s2i + τ 2 ). Here, β is E[βi ] and τ 2 captures between-site heterogeneity.
Estimating τ 2 : Following DL, we use a method-of-moments estimator based on Cochran’s
Q statistic computed under fixed-effect weights wi ≡ 1/s2i . Defined as
Pk k k Pk
wi β̂i X 2 X wi2
β̂ ≡ Pi=1
k
, Q≡ wi β̂i − β̂ , C≡ wi − Pi=1
k
.
i=1 w i i=1 i=1 i=1 wi
With df = k − 1, the DL estimator is
2 Q − df
τ̂DL = max 0, .
C
2
Pooled Effect, Standard Error, and Inference: Given τ̂DL , define random-effects
inverse-variance weights
1
wi∗ ≡ 2 2
.
si + τ̂DL
The pooled estimator of β is the weighted mean
Pk k
!−1/2
∗
i=1 wi β̂i
X
β̂ = Pk ∗
, SE β̂ = wi∗ .
i=1 wi i=1
Testing for heterogeneity: Cochran’s Q tests homogeneity and is approximately χ2df
under H0 : τ 2 = 0, allowing us to test the null of no heterogeneity in effects across sites. We
report the test statistic and p-value for each outcome.
34
See also Jackson and Mackevicius (2024) and Jackson (2025), who implement similar analyses.
84
Appendix Table F.5
Testing for cross-site heterogeneity in survey outcomes
Meta-analysis
outcome β̂ SE(β̂) p-value τ̂ 2 Q Q p-value
All Rent Paid 0.070 0.017 >0.000 0.0000 1.785 0.618
All Bills Paid -0.003 0.017 0.883 0.0000 1.894 0.595
COVID Positive -0.031 0.021 0.139 0.0007 5.455 0.141
Feeling Anxious -0.035 0.023 0.133 0.0009 5.063 0.167
Worried About Eviction -0.037 0.014 0.010 0.0000 2.514 0.473
Experienced Homelessness -0.016 0.010 0.096 0.0000 0.898 0.826
Stayed in a Homeless Shelter -0.002 0.004 0.716 0.0000 5.288 0.152
Was Food Insecure -0.026 0.016 0.101 0.0000 1.877 0.598
Notes: This table reports the output from a random-effects model designed to test for heterogeneity across sites following
Jackson (2025) using DerSimonian and Laird (1986) for survey outcomes. Column one reports m̂u reports the pooled effect
across sites, while columns 2 and 3 report the standard error and p-value for the pooled effect. Column four reports τ̂ 2 , the
estimate of the variance in true site-specific effects. Columns five and six report Cochran’s Q and its p-value under the null
that τ 2 = 0.
F.2.2 Site heterogeneity-analysis results
Tables F.5–F.9 report the results of our analysis of cross-site heterogeneity. For all survey
outcomes, we fail to reject the null that τ 2 = 0, with the strongest evedince of cross-site
heterogeneity for the “COVID Positive,” “Feeling Anxious,” and “Stayed in a Homeless
Shelter” outcomes. We similarly fail to reject the null for short-run financial health outcomes
and short- and long-run address mobility. For longer-run financial health outcomes we reject
the null for Credit Scores and on Any Personal Bankruptcy. Overall, we interpret our findings
from this analysis as providing little evidence of site-specific heterogeneity, except for some
indication that sites may have differed in their impacts on longer-term financial health. The
null findings may be partly driven by the modest number of sites in our study.
85
Appendix Table F.6
Testing for cross-site heterogeneity in financial health outcomes (2 months)
Meta-analysis
outcome β̂ SE(β̂) p-value τ̂ 2 Q Q p-value
Credit Score 0.514 1.166 0.659 2.3010 6.187 0.186
Balance Across all Trades ($1,000s) -0.020 0.477 0.966 0.2441 5.043 0.283
Balance in Collections ($1,000s) -0.034 0.034 0.321 0.0021 6.369 0.173
Balance in Utility Collections ($1,000s) -0.011 0.007 0.133 0.0000 3.392 0.494
Balance in Delinquent Accounts ($1,000s) -0.136 0.134 0.308 0.0000 1.730 0.785
Any Auto Loan or Lease -0.000 0.005 0.991 0.0000 1.505 0.826
Any Personal Bankruptcy -0.000 0.001 0.870 0.0000 4.718 0.317
Any Open Revolving Line of Credit 0.004 0.006 0.498 0.0000 5.640 0.228
Notes: This table reports the output from a random-effects model designed to test for heterogeneity across sites following
Jackson (2025) using DerSimonian and Laird (1986) for short-run financial health outcomes. Column one reports m̂u reports
the pooled effect across sites, while columns 2 and 3 report the standard error and p-value for the pooled effect. Column four
reports τ̂ 2 , the estimate of the variance in true site-specific effects. Columns five and six report Cochran’s Q and its p-value
under the null that τ 2 = 0.
Appendix Table F.7
Testing for cross-site heterogeneity in financial health outcomes (10 months)
Meta-analysis
outcome β̂ SE(β̂) p-value τ̂ 2 Q Q p-value
Credit Score -1.714 2.008 0.393 10.7796 9.839 0.043
Balance Across all Trades ($1,000s) -0.258 0.630 0.682 0.0000 2.242 0.691
Balance in Collections ($1,000s) 0.013 0.034 0.689 0.0000 0.297 0.990
Balance in Utility Collections ($1,000s) -0.012 0.010 0.229 0.0000 0.841 0.933
Balance in Delinquent Accounts ($1,000s) 0.053 0.154 0.728 0.0000 2.488 0.647
Any Auto Loan or Lease -0.002 0.007 0.812 0.0000 0.806 0.938
Any Personal Bankruptcy -0.003 0.004 0.478 0.0000 12.419 0.014
Any Open Revolving Line of Credit -0.008 0.007 0.219 0.0000 3.952 0.413
Notes: This table reports the output from a random-effects model designed to test for heterogeneity across sites following
Jackson (2025) using DerSimonian and Laird (1986) for longer-run financial health outcomes. Column one reports m̂u reports
the pooled effect across sites, while columns 2 and 3 report the standard error and p-value for the pooled effect. Column four
reports τ̂ 2 , the estimate of the variance in true site-specific effects. Columns five and six report Cochran’s Q and its p-value
under the null that τ 2 = 0.
86
Appendix Table F.8
Testing for cross-site heterogeneity in mobility (2 months)
Meta-analysis
outcome β̂ SE(β̂) p-value τ̂ 2 Q Q p-value
Change of Address Infutor -0.001 0.002 0.714 0.0000 0.487 0.975
Change of Address Experian -0.003 0.006 0.539 0.0000 1.712 0.788
Notes: This table reports the output from a random-effects model designed to test for heterogeneity across sites following
Jackson (2025) using DerSimonian and Laird (1986) for short-run residential mobility outcomes. Column one reports m̂u
reports the pooled effect across sites, while columns 2 and 3 report the standard error and p-value for the pooled effect.
Column four reports τ̂ 2 , the estimate of the variance in true site-specific effects. Columns five and six report Cochran’s Q and
its p-value under the null that τ 2 = 0.
Appendix Table F.9
Testing for cross-site heterogeneity in mobility (10 months)
Meta-analysis
outcome β̂ SE(β̂) p-value τ̂ 2 Q Q p-value
Change of Address Infutor -0.016 0.005 0.003 0.0000 1.280 0.865
Change of Address Experian 0.006 0.009 0.525 0.0000 1.768 0.778
Notes: This table reports the output from a random-effects model designed to test for heterogeneity across sites following
Jackson (2025) using DerSimonian and Laird (1986) for longer-run residential mobility outcomes. Column one reports m̂u
reports the pooled effect across sites, while columns 2 and 3 report the standard error and p-value for the pooled effect.
Column four reports τ̂ 2 , the estimate of the variance in true site-specific effects. Columns five and six report Cochran’s Q and
its p-value under the null that τ 2 = 0.
87
File and source
- File
- nber-w32463.pdf
- Size
- 1,534,410 bytes
- SHA-256
- f7cd88a1a21dae18105ddaa29d7d7447582611016faab9d77787cc362816fc18
- Our copy
- nber-w32463.pdf
- Original
- www.nber.org