Eviction and Poverty in American Cities

Abstract

More than two million U.S. households have an eviction case filed against them each year. Policy makers at the federal, state, and local levels are increasingly pursuing policies to reduce the number of evictions, citing harm to tenants and high public expenditures related to homelessness. We study the consequences of eviction for tenants using newly linked administrative data from two major urban areas: Cook County (which includes Chicago) and New York City. We document that before housing court, tenants experience declines in earnings and employment and increases in financial distress and hospital visits. These pre trends pose a challenge for disentangling correlation and causation. To address this problem, we use an instrumental variables approach based on cases randomly assigned to judges of varying leniency. We find that an eviction order increases homelessness and hospital visits and reduces earnings, durable goods consumption, and access to credit in the first two years. Effects on housing and labor market outcomes are driven by effects for female and Black tenants. In the longer run, eviction increases indebtedness and reduces credit scores.

I. Introduction

More than two million eviction court cases are filed in the United States each year. These cases predominantly involve low-income and minority households. About half of proceedings end in a court order for eviction: a judgment requiring the tenant to vacate the property.¹ According to data collected by the Organisation for Economic Co-operation and Development (OECD), the United States is an outlier in the number of eviction cases per renter household, with a rate 1.5 times higher than the next highest country (Canada) and at least 3.8 times higher than the remaining 10 countries for which data are available (OECD 2020). In recent years, policy makers at the federal, state, and local levels have introduced assistance programs and legislative changes aimed at reducing the number of evictions, frequently citing harms to tenants and the high costs of homelessness-related public services.² Measuring the consequences of an eviction for tenants is crucial for evaluating these reforms and, more broadly, for understanding the role of housing instability as a driver of poverty and inequalities in income, socioeconomic mobility, and health that have been documented in recent literature (Piketty and Saez 2003; Chetty et al. 2014; Case and Deaton 2015).

Despite the large number of tenants who interact with housing courts in the United States each year and the growing interest from policy makers, the consequences of eviction for households are not well documented or understood. While researchers have argued that eviction is a cause of poverty, homelessness, poor health, and other forms of physical and material hardship (e.g., Desmond 2012, 2016), quantitative empirical research in this area has been hampered by two main challenges. First, it is difficult to link data on households facing eviction to data on their subsequent outcomes. Second, it is not obvious how to separate the effect of eviction from the impact of correlated sources of distress such as job loss or declining health. This article overcomes these barriers to provide new evidence on the effect of eviction on earnings, employment, residential mobility, interactions with homelessness services, financial distress, and health. We link newly constructed data sets based on housing court records from two large urban areas—New York City, NY, and Cook County, IL (which includes the city of Chicago)—to a broad range of administrative data sets. These linked data allow us to document and characterize tenants’ outcome trajectories several years before and after their eviction case. To identify the causal impact of the eviction order, we use an instrumental variables (IV) research design that relies on the random assignment of cases to judges who systematically vary in their tendency to evict.³

We first show that tenants in our linked housing court sample differ substantially from randomly chosen tenants who live in the same neighborhoods. Compared to these neighbors, tenants we observe in housing court have lower earnings, lower employment, less access to credit, and more debt in collections. In addition, both evicted and nonevicted tenants experience striking drops in earnings, employment, and credit scores and rising hospital visits, unpaid bills, and payday loan inquiries in the two years before the case. These “Ashenfelter dips” are more pronounced for evicted tenants and suggest the presence of unobserved factors that are correlated with both the eviction decision and postcourt outcomes, and are likely to introduce bias in estimates based on cross-sectional or difference-in-differences comparisons. For this reason, our main estimates are based on a quasi-experimental IV research design using the random assignment of judges.

With the IV approach, we find that eviction causes spikes in homelessness and increases in residential mobility. In the first year after case filing, an eviction order increases the probability of observing the tenant at a new address by 8 percentage points (28% of the nonevicted mean) and increases the probability of staying in emergency shelters by 3.4 percentage points (more than 300% of the nonevicted mean). The effects on residential mobility and homelessness persist through the second year after filing. Yet these increases in housing instability do not result in large changes in neighborhood quality: after the court case, evicted tenants live in neighborhoods with similar poverty rates as tenants who are not evicted. This finding suggests that it is unlikely that the effects on other outcomes arise due to changes in neighborhood environment, as in studies of housing mobility programs (Chetty, Hendren, and Katz 2016) or public housing demolitions (Chyn 2018).

During the two-year period of increased housing instability and homelessness, eviction negatively affects earnings. Our IV estimates imply that eviction lowers earnings in the year after filing by 323 per quarter (8% of the nonevicted mean), which is similar to evicted tenants’ average drop in quarterly earnings in the year leading up to case filing (337). The effect on earnings is larger in the second year after the case, with eviction causing a 613 (14%) reduction in quarterly earnings. The estimated effects on employment are more modest, with eviction causing a 1.5 percentage point reduction in the fraction of quarters employed in the year after the case, and a 1.8 percentage point reduction two years after the case, neither of which is statistically significant at the 10% level. The labor market effects of eviction are largely concentrated in the two years after filing. We find particularly sharp negative effects for female and Black tenants, who drive the effects on labor market outcomes, residential mobility, and interactions with homelessness services. This pattern is consistent with ethnographic research that suggests eviction may have a larger effect on women (Desmond 2012, 2016; Desmond et al. 2013) and with research that finds that Black households experience discrimination while searching for housing (Bayer et al. 2017; Christensen and Timmins 2022).

Eviction also worsens financial health and credit access during and beyond the initial period of increased housing instability and homelessness. Using data from linked credit reports, we find that eviction causes reductions in a composite index of financial health of roughly 0.1 std. dev. in the first and second years after the case filing, by 0.21 std. err. three to four years after filing, and by 0.26 std. dev. five to six years after filing. The declines are driven by increases in debt and lower credit scores.⁴ We find evidence that eviction reduces the likelihood of having an automobile loan or lease, which may be viewed as a proxy for durable-goods consumption (Dobkin et al. 2018; Agarwal et al. 2023). The effects on credit scores of 16.5 points in the second year after the case are similar in magnitude to the effect of removing a bankruptcy flag (Gross, Notowidigdo, and Wang 2020; Dobbie et al. 2020).

Finally, we find that eviction increases the number of hospital visits in the year after court filing by 0.19 visits (29%) and increases visits for mental health–related conditions during the same period by 0.05 visits (133%). The timing of these effects coincides with the disruptions to tenants’ housing circumstances in the year after filing.

Our analysis is relevant for ongoing policy debates. First, we find that eviction causes significant disruptions that are reflected in increases in residential mobility, homelessness, and hospital use; reductions in earnings; and sustained damage to credit records. These costs are key inputs to the evaluation of a range of policies, such as emergency rental assistance, legal aid to tenants facing eviction, and, most directly, making eviction proceedings more lenient toward tenants.⁵ Given the large social costs of homelessness (Evans, Philips, and Ruffini 2019), our finding that a court-ordered eviction increases the likelihood of emergency shelter use suggests a role for policy in the eviction court setting to reduce homelessness. Second, we show that eviction is frequently preceded by adverse events, which may reflect the inadequacy of existing social insurance policies or self-insurance in preventing evictions. Third, we find that the effects of eviction are driven by traditionally vulnerable groups: Black and female tenants. Since these groups also tend to be overrepresented in eviction proceedings, policies aimed at averting eviction may especially benefit them.

This article is related to a sizable literature in sociology that studies eviction of low-income renters (Desmond 2012; Desmond, Gershenson, and Kiviat 2015; Desmond and Gershenson 2016, 2017; Desmond 2016). Our work builds on and extends this literature in several ways. First, we show that the research designs used in previous work on evictions may be vulnerable to selection bias. Second, to address this selection bias, we use a quasi-experimental research design to estimate the causal effects of eviction by leveraging the random assignment of judges to eviction cases. Third, we create a novel data set of eviction court records linked to administrative data, which helps mitigate the concerns that may arise when using survey data, including selective nonresponse and misreporting (Meyer, Mok, and Sullivan 2015; Dutz et al. 2021). The linked data also let us characterize tenants’ housing, labor market, health, and credit circumstances in the lead-up to and aftermath of filing. Finally, we provide a unified analysis across two large U.S. urban areas not previously studied using more than a decade of administrative data, lending support to the external validity of our findings.

We examine the impact of eviction on earnings, homelessness, and financial health, outcomes that have not been studied in prior work. We find that eviction causes increases in homelessness and reduces earnings in the two years after the case filing and leads to longer-run deterioration in financial health. Prior studies have examined the effect of eviction on loss of employment (Desmond and Gershenson 2016), mental health (Desmond and Kimbro 2015), and moves to high-poverty neighborhoods (Desmond and Shollenberger 2015). Relative to these studies, using our quasi-experimental research design, we find more modest effects of eviction on employment and no effect on the poverty rate of neighborhoods to which evicted tenants move.

While there is relatively little work on eviction in economics, related work examines the effect of homeowners’ foreclosure on health outcomes (Currie and Tekin 2015), subsequent homeownership, housing and neighborhood conditions (Molloy and Shan 2013), and credit scores (Brevoort and Cooper 2013). A related study by Diamond, Guren, and Tan (2020) examines the effect of foreclosure on residential mobility, homeownership, divorce, measures of neighborhood quality, and credit reports using a randomized-judge design. As part of their analysis, Diamond, Guren, and Tan (2020) consider the effect of a landlord’s foreclosure on tenants. We view our work as complementary, since eviction and foreclosure are different court processes and affect different populations.⁶ We consider several additional dimensions that eviction is likely to affect, including employment, earnings, homelessness, and hospital use.

Last, our work is related to studies of the incidence and drivers of eviction filings. In particular, several recent studies analyze the effects of expansions of and cuts to the social safety net on eviction rates. Gallagher, Gopalan, and Grinstein-Weiss (2019) find that expansions of ACA Marketplace subsidies substantially reduced eviction filing rates, and Zewde et al. (2019) find that Medicaid expansions were associated with reductions in county-level filing rates and eviction rates. Fetzer, Sen, and Souza (2020) study the effect of cuts to rental subsidies in the United Kingdom and find that these substantially increased rental arrears and evictions. These results are consistent with our findings that adverse health, labor market, and credit outcomes precede and may contribute to appearing in housing court and being evicted. Additional risk factors are documented in Desmond et al. (2013), who point to children as a risk factor for eviction, and Desmond and Gershenson (2017), who find that family size, job loss, neighborhood crime, and network disadvantage are additional risk factors.

The remainder of the article is organized as follows. Section II provides institutional details relevant for understanding the eviction process in Cook County and New York. Section III describes the data collection and record linkage processes. Section IV describes our samples, provides new descriptive evidence on the evolution of outcomes among evicted and nonevicted tenants around a court filing, and explores selection into eviction. Section V formalizes our empirical framework and tests the key underlying assumptions. Section VI presents the main results of our analysis. Section VII concludes. All appendix material can be found in the Online Appendix.

II. Institutional Context

This section describes the legal process of eviction and other relevant institutional details. In Cook County and New York, as in most jurisdictions, the housing court process begins with a notice served to the tenant by the landlord, followed by one or more court hearings, and finally a judge’s decision on whether to issue an eviction order that requires the tenant to vacate the property.

A landlord must serve the tenant a written notice to begin the eviction court process. The notice typically includes the reason for terminating the lease and the number of days until termination. A landlord may seek an eviction for any alleged violation of the lease terms, and nonpayment of rent is the most commonly stated reason.⁷ In both Cook County and New York, the landlord has no discretion over the district that will handle their case, since the district is determined by the address of the property under dispute. As we discuss, cases in both jurisdictions are randomly assigned to courtrooms, with judges assigned to courtrooms on a fixed rotational basis.

Nearly all eviction cases are handled in a resolution process overseen by a judge.⁸ When the landlord and tenant meet in a courtroom, the hearing is typically brief: court observation studies have found that the average eviction hearing lasts only a few minutes (Doran et al. 2003). Tenants are usually unrepresented, while landlords are usually represented by an attorney.⁹

To proceed with an eviction, the landlord needs a court order that authorizes the enforcement agent, such as a Sheriff or marshal, to execute the eviction order. In both jurisdictions, we define an eviction as a case ending with an eviction order. This definition is based on whether the last recorded outcome in the case history provides legal authority for the landlord to take possession of the property via an enforcement agent.¹⁰  Online Appendix C.C explains in more detail how we construct eviction orders from the housing court data. In cases where the landlord is seeking rental arrears, the judge may include an order to pay rental arrears along with the eviction order, called a money judgment.

The alternative to an eviction order is often a formal agreement between the landlord and tenant that is approved by the judge. Such agreements typically include a payment plan, and they may also set terms for continued occupancy of the unit.¹¹ The landlord may return to court to pursue an eviction order if the tenant doesn’t satisfy the terms of an initial agreement. Cases can be discontinued, which happens if the landlord decides not to pursue the case further. Only 5% of nonevictions are dismissals that bar the landlord from bringing another eviction case with the same allegations against the tenant.

An eviction order may or may not be followed by the execution of the order by an enforcement officer such as a sheriff or marshal. We refer to the execution of an eviction order as an enforcement, and it typically involves changing the locks and removing the tenant’s possessions. Whether an eviction order is enforced depends on several factors. For example, the landlord may choose not to file the order with the enforcement agent because they must pay an additional fee. The landlord and tenant may also come to an informal agreement. Finally, the tenant may choose to vacate the unit before a sheriff or marshal is scheduled to enforce the eviction order, in which case the landlord may cancel the enforcement of the order.

There are several reasons an eviction order may affect tenants’ future outcomes. First, an eviction order legally obligates a tenant to move, either following or in anticipation of the enforcement of the order, and thus to incur the costs associated with searching for new housing, relocating, and reorienting the household’s work and schooling arrangements. Second, eviction orders and filings are public records in most jurisdictions, and an order can also be recorded as a civil judgment on the tenant’s credit report. Eviction filings and eviction orders are commonly used in background screenings by landlords, employers, and creditors, and therefore an eviction can make it harder for tenants to secure future rental contracts, employment, or loans. Finally, in cases where the landlord seeks a money judgment, an eviction order will typically include a money judgement, which can be used by the landlord to obtain an order for garnishment of wages, tax refunds, or other assets. Garnishment requires a separate court process and is rare in practice. See Online Appendix B for additional institutional details.

II.A. Cook County

Roughly 33,000 eviction cases are filed in Cook County each year. These are handled by the Forcible Entry and Detainer Section of the Circuit Court of Cook County. Roughly 80% of Cook County cases are joint action cases, where the landlord is seeking payment of rental arrears in addition to possession of the property. The remaining 20% of cases are single action cases, where the landlord is only seeking possession of the property. The court divides the county into six districts, each with its own courthouse, eviction courtrooms, and eviction case judges. Landlords must file eviction cases in the district in which the property is located. The city of Chicago is located entirely within Cook County, IL, and eviction cases filed in the city represent about 75% of the county’s case volume.

Eviction cases are randomly assigned to courtrooms within a district by a computer algorithm. Judge assignments to courtrooms are set in advance, and therefore random assignment to a courtroom is effectively random assignment to a judge.

Approximately 65% of eviction cases in Cook County end with an eviction order. We estimate the share of nonevicted cases with a formal agreement to be upward of 39%.¹² Around 45% of cases without an eviction order in Cook County are discontinued, and roughly 5% are dismissed. The Cook County Sheriff’s Office executes about 26% of cases ending with an eviction order.¹³

II.B. New York City

Each year, around 240,000 cases are filed in housing court in New York. The Civil Court of New York City, part of the state Unified Court System, oversees the New York City Housing Court. Housing Court hears cases involving landlord-tenant disputes or housing code violations. Cases are handled by seven courthouses: one for each county (borough) in New York City (Bronx, Kings, New York, Queens, and Richmond Counties) and two smaller, specialized courts in Harlem and Red Hook. The courthouse is determined by location of the filing address. The vast majority of eviction cases heard in housing court are nonpayment filings (86%), with the remaining being other lease violation disputes known as “holdover” cases (14%).

Cases are randomly assigned to courtrooms by the Housing Court Information System (HCIS) computers in the courthouse of the assigned case. Judges rotate through courtrooms for year-long terms on a predetermined rotation system. Cases are assigned to courtrooms rather than judges, and therefore if the judge rotates out of a courtroom during an active case, the case will remain in the assigned courtroom. Some types of cases, such as those involving the public housing authority, are not randomly assigned to courtrooms, and we exclude these from the analysis. For details, see Online Appendix C.B.

In New York, about 35% of nonpayment cases end with an eviction order. Among those ending without an eviction order, approximately 64% end with a settlement agreement, 29% are discontinued, and 5% are dismissed. The enforcement of an eviction order is conducted by a city marshal. In our data, 31% of cases ending with an eviction order in New York result in an enforcement of the order conducted by a city marshal.

III. Data and Linkage

Our empirical analysis uses court records from Cook County, IL, and New York, NY, linked to administrative data sets measuring earnings and employment, residential address histories, interactions with the homelessness services system, and credit bureau records. We also link the New York court data to records of hospital visits. This section summarizes our data sources, sample construction, data linkage, and main outcomes. We provide additional details in Online Appendix C.

III.A. Court Data

Our linked data sets are based on the near universes of court records for Cook County for 2000–2016 and for New York for 2007–2016. Each court record includes the residential address of the disputed housing unit and the names of one or more tenants. The unit of analysis is the case-individual, so each tenant who appears as a defendant in the case will have a separate record.¹⁴ Other key elements we observe in the court records are case type, filing date, courtroom and date assignment, name of the landlord, attorneys’ names, the amount claimed by the landlord (ad damnum amount), and whether an eviction order was granted.¹⁵ We observe other judge decisions throughout the case, such as whether to grant a continuance in the case or a stay of the eviction order. We define an eviction as a case ending with an eviction order.¹⁶

While the data are similar across our two settings, there are a couple of differences to note. In Cook County, the data include the value of any money judgment awarded and the name of the judge associated with each action in the court record, but we do not observe either in New York.

1. Sample Restrictions

We impose several restrictions on our court samples. In both locations, we drop eviction cases associated with businesses, cases associated with condominiums, cases with a missing defendant name or address, cases involving more than 100,000 in claimed damages, and cases filed during a week in which only a single judge (courtroom in New York) was hearing cases. These sample restrictions are necessary to focus our analysis on residential eviction cases involving renters where we can link to outcomes and construct the instrument. We also restrict the sample to cases in which the judge (courtroom in New York) presided over a minimum number of cases during the year: 100 in Cook County and 500 in New York. This restriction removes judges/courtrooms that hear substantially fewer cases than is typical in the setting, which removes noise in the instrument.¹⁷

In New York, some cases are not randomly assigned to courtrooms: cases involving public housing units, cases involving co-ops or condominiums, cases assigned based on ZIP code through several policy initiatives, cases for family members of active military personnel, and cases involving the district attorney’s office or the New York City Police Department. We can identify these cases directly in the New York courts data and drop them from our sample.

The court sample includes around 414,000 cases for Cook County and 580,000 cases for New York before linking to outcomes data. Online Appendix Tables C.1 and C.2 describe how sample counts change with these restrictions in Cook County and New York, respectively.

III.B. Outcomes Data

We link the court records to multiple administrative data sets. Below, we describe these data sets and define the outcomes we study in our analysis. We separately analyze linked records for Cook County and New York because of data security restrictions.¹⁸ Additional details on data linkage and sample construction are provided in Online Appendix C.

1. Earnings and Employment

In both settings, we measure earnings and employment using quarterly records derived from state unemployment insurance (UI) data systems. Our main earnings outcome is quarterly wage earnings, and our main employment outcome is an indicator for positive earnings in a given quarter. We restrict the analysis to tenants who are 18 to 55 years old at the time of case filing to exclude individuals aging into retirement. Earnings and all other dollar amounts are expressed in 2016 US using the CPI-U for the two metropolitan areas we study. Employment and earnings records only cover formal employment and exclude individuals not covered by UI benefits, such as the self-employed.

The UI records for New York are from the New York State Department of Labor and do not include states other than New York. They cover the years 2004 to 2016. The UI records for Cook County are from the Longitudinal Employer-Household Dynamics (LEHD) Employer History File, a restricted Census Bureau data set (see Abowd, Haltiwanger, and Lane 2004; Vilhuber 2018, for more details on the LEHD). We measure employment using the LEHD file that contains a flag for any positive earnings in any of the 50 states or the District of Columbia. We observe quarterly earnings for Illinois, the District of Columbia, and 11 other states for which we were granted access to earnings data.¹⁹ The years available vary by state, but all states have data from 1995 to 2014.²⁰

2. Residential Mobility

In Cook County, our primary data source for measuring residential address changes is the Census Bureau’s Master Address File Auxiliary Reference File (MAFARF), which provides addresses of residence and associated census geographic identifiers by year.²¹ We use the MAFARF to build an indicator for the tenant being observed at the filing address in each time period. While the data are rarely missing, some individuals do not have an address listed in certain years. Online Appendix Figure E.1 plots the proportion of evicted and nonevicted tenants with address information each year, relative to the filing year. Evicted tenants are somewhat less likely to have a reported address, and this difference grows moderately after the eviction case is filed. When studying mobility outcomes below, we also report and discuss robustness results based on how these missing observations are handled.

In New York, we combine two sources of address histories: consumer reference data from Infutor Data Solutions and administrative benefits records.²² Similar to the Cook County data, we define a tenant as not at their eviction address if we observe them at a different address than the one listed on the court filing according to either the benefits data or the Infutor data in the relevant outcome window. A concern with the New York sources of address data is that the availability of address information could be affected by an eviction. However, Online Appendix Table C.3 shows that eviction is only weakly correlated with the probability of having an address from the Infutor data or the benefits data. Online Appendix Table I.3 shows that estimates of the effect of eviction on residential mobility in New York are not particularly sensitive to using either data source on its own in cases when both are available.

Measuring address-level moves at an annual frequency in the United States is challenging, particularly so for our population of unstably housed tenants. We believe these administrative data sets provide the best measures available.

Using the address data described already, we link to neighborhood poverty rates. In Cook County, we use census tract–level neighborhood poverty rates constructed from restricted-access American Community Surveys (ACS) from 2005 to 2018, based on five-year moving averages. In New York, we use the publicly available census tract five-year estimates from the ACS 2006–2010.

3. Homelessness

We measure interactions with homelessness services in both settings using local Homeless Management Information System (HMIS) data.²³ The Cook County HMIS database is managed by All Chicago and is similar to the data set used in Evans, Sullivan, and Wallskog (2016). The HMIS records are linked to census identifiers and are studied in the Census Research Data Center. They capture the years 2014 to 2018 and include individual-level data on stays in emergency shelters as well as other interactions with homelessness prevention services. Similarly, the HMIS data in New York capture individual-level applications to and stays in the city’s vast shelter system, as well as diversions through homeless prevention programs. These data come from the New York City Department of Homeless Services and cover 2003 to 2017. We use these data to construct two outcomes: an indicator for the individual staying in an emergency shelter, and an indicator for the individual interacting with any homelessness services. In Cook County, homelessness services include emergency shelter use, permanent supportive housing, coordinated assessment of need, rapid rehousing, transitional housing, and street outreach. In New York, this indicator additionally includes applications to shelter, which cover instances where families are diverted or deemed ineligible.²⁴

4. Financial Health

We measure financial health with credit records from Experian, one of the three major credit bureaus in the United States.²⁵ For Cook County, the linked credit report data are biennial snapshots from March 2005 to March 2017 and an additional snapshot for September 2010. For New York, the linked credit report data are quarterly snapshots from June 2014 to September 2019. For both locations, we measure overall financial health using VantageScore 3.0, which is on a scale of 300–850; scores under 600 are considered subprime. We measure unpaid bills as the total balance currently either 30 days or more delinquent or in collections, where the latter are balances that the lender turns over to a collections agency following a period of delinquency, typically at least 30 days. We construct an indicator for any positive balance on an auto loan or lease, which has been used as a proxy for durable-goods consumption (Dobkin et al. 2018; Agarwal et al. 2022). We measure whether the tenant has no open source of revolving credit, such as a credit card, which serves as a proxy for having limited access to credit.

As a summary measure, we create an index of financial health based on the credit bureau variables described above and following the approach of Dobbie, Goldsmith-Pinkham, and Yang (2017). Each component of the index is standardized based on the nonevicted mean and standard deviation in the filing year. We sum the standardized components, with the indicator for no revolving credit and the amount of unpaid bills entering the index negatively, so that all components can be viewed as contributing to financial health. We then restandardize the index based on the mean and standard deviation of the index for the nonevicted group in the filing year. Last, we observe payday loan account inquiries and borrowing for individuals in Cook County, which includes both online and storefront loans. The majority of these loans are originated online. We describe the payday loans data in detail and present the analysis in Online Appendix C.G.

5. Health

For New York, we also measure health outcomes using data from the New York State Department of Health’s Statewide Planning and Research Cooperative System. This data set includes all inpatient and outpatient (including Emergency Department) hospital visits in New York State from 2004 to 2016.²⁶ For each hospital visit, the data include the date of intake and a primary diagnosis code (ICD-9 code). We focus on the total number of (nonpregnancy-related) hospital visits, including inpatient or outpatient visits, the total number of emergency department visits, and the total number of hospital visits for mental health conditions.²⁷

III.C. Data Linkage

We link court records to other administrative data sets using tenant names and addresses. To link Cook County court records to Census Bureau–held data sets, the Census Bureau used names and addresses to link individuals to their unique PIK.²⁸ The PIK rate for the Cook County sample is 52%. PIKs are then used to link to other restricted data sets held in the Census Bureau Research Data Centers.

To link New York court records to outcomes, we first use names and addresses to link individuals to historical benefits data from the New York City Human Resource Administration for 2004 to 2016. The data include individuals receiving Medicaid, SNAP, TANF, or other city-specific cash subsidies. Online Appendix D.B describes this process in detail. The data have personal identifiers, including Social Security Number and date of birth, that we use to link individuals to the outcomes data. The data also include demographic information such as age, gender, race, and ethnicity. The benefits data capture roughly two million unique New Yorkers each year. Because receiving benefits may be endogenous to the eviction court outcome, we restrict the sample to court records that match the benefits data prior to an eviction filing. Roughly 40% of the court records have a match in the benefits data. Individuals in the linked data have somewhat lower incomes and are more likely to be older, female, and have children when compared to the overall population in housing court (NYC Office of Civil Justice 2016).

Last, we link court records to measures of financial health from Experian credit reports. This linkage yields match rates of 61.3% in Cook County and 68% in New York, which are comparable to match rates in previous studies that link data to records from the major credit bureaus.²⁹ The linked credit sample consists only of individuals who have a credit record. In low-income neighborhoods, more than 70% of all adults have credit records (Brevoort et al. 2015).

Online Appendix D compares the court record populations in Cook County and New York to the subpopulations successfully linked to outcomes and also examines court record characteristics predictive of a match. More disadvantaged tenants (those without legal representation or those evicted) are somewhat less likely to be linked to census records in Cook County and slightly more likely to be linked to benefits data in New York. The pattern is similar for links to Experian data in Cook County but the opposite for links to Experian data in New York. These patterns will not affect the internal validity of our results since, conditional on linking to outcomes, the baseline characteristics of the case and tenant are not predictive of judge stringency, as we show in Section V.

Online Appendix D also studies the relationship between judge stringency and the probability of being linked to outcome data. We find that in three out of the four analysis samples, judge stringency is uncorrelated with the probability that a case is linked to outcomes. The exception is the Cook County linked census sample, which has an economically small but statistically significant relationship between stringency and the probability of being assigned a PIK (Online Appendix Table D.1). Moving from the 10th percentile of stringency to the 90th percentile of stringency—a 7 percentage point difference—is associated with only a −0.38 percentage point reduction in the likelihood of having a PIK (−0.054 × 0.07, using the estimate from Online Appendix Table D.1, column (2)). This correlation likely arises due to the census linkage process, which may incorporate postfiling information that is impacted by eviction.³⁰ We emphasize that the correlation between stringency and the probability of having a census PIK does not threaten the internal validity of our estimates because conditional on having a PIK, judge stringency is unrelated to individual and case characteristics, which we discuss below and show in Table III. We also show in Online Appendix Table G.4 that stringency is uncorrelated with lagged values of all our outcomes that are linked using census PIKs. Last, there is no relationship between stringency and being linked to the New York benefits sample, which yields a similar pattern of results to the Cook County linked census sample, suggesting differences in PIK rates are not driving the effects we document in Section VI.

TABLE III.

Testing Balance

	Cook County		New York
	Evicted	Stringency	Evicted	Stringency
	(1)	(2)	(3)	(4)
Age at case	−0.03329***	−0.00012	−0.00403***	−0.00001
	(0.00376)	(0.00020)	(0.00016)	(0.00001)
Female	0.00882	0.00041	−0.04413***	−0.00009
	(0.00644)	(0.00036)	(0.00310)	(0.00011)
Black	0.06297***	0.00012	0.00923***	0.00010
	(0.00628)	(0.00028)	(0.00323)	(0.00018)
White	0.00358	0.00011	−0.01494**	−0.00032
	(0.00582)	(0.00030)	(0.00616)	(0.00027)
Hispanic	0.05957***	0.00045	−0.00743**	0.00001
	(0.00603)	(0.00030)	(0.00368)	(0.00017)
Neighborhood poverty rate (5 yr. avg.)	0.5540***	0.00208	−0.02487*	−0.00025
	(0.04813)	(0.00221)	(0.01453)	(0.00066)
Ad damnum (in 1,000s)	0.00731***	0.00001	0.00001***	−0.00000
	(0.00055)	(0.00002)	(0.00000)	(0.00000)
No prior case	−0.04037***	−0.00013	−0.01228***	−0.00014
	(0.00221)	(0.00013)	(0.00413)	(0.00014)
Joint action	0.01183**	−0.00061**
	(0.00525)	(0.00025)
Observations	301,000	268,000	150,662	150,662
Joint F-Statistic	102.3	1.497	224.8	1.007
p-value	.000	.104	.000	.443

Notes. For each location, the left column presents results from a regression of eviction on case and defendant characteristics, and the right column shows results from a regression of judge stringency on case and defendant characteristics. Neighborhood poverty rate is the five-year average poverty rate in the defendant’s census tract. Ad damnum is the amount the landlord listed as owed by the defendant at the time of filing. Joint action is an indicator for the case type in which the landlord is seeking both an eviction order and a money judgment rather than only an eviction order and is specific to Cook County. No prior case is an indicator for the defendant having no prior eviction case in our sample. All regressions include indicators for each right-side variable having a missing value, which are not reported in the table. All regressions include court-year-quarter fixed effects. Standard errors are shown in parentheses and are clustered at the judge(courtroom)-year level. Cook County observation counts are rounded in accordance with U.S. Census Bureau disclosure requirements. Cook County results were approved for release by the U.S. Census Bureau, authorization no. CBDRB-FY22-072. * p < .1, ** p < .05, *** p < .01.

IV. Trends and Evidence of Selection

This section provides new descriptive facts about the demographics, earnings, employment, housing, health, and financial circumstances of tenants in housing court, based on the linked panel data described in the previous section. We show that while evictions primarily occur in neighborhoods with high poverty rates, tenants in our linked housing court sample are also negatively selected on precourt earnings and employment relative to randomly chosen renters who live in the same neighborhoods. Within housing court, we also find notable differences between evicted and nonevicted tenants. These differences show up in both levels and trends leading up to the moment the case is filed for nearly all outcomes considered, suggesting the presence of unobserved factors that are correlated with both the eviction decision and postcourt outcomes. This motivates our IV research design described in Section V.

IV.A. Tenants in Housing Court

Figure I maps the location of evictions in 2010 by census tract for Cook County and New York, together with tract-level poverty rates. Although evictions occur throughout both areas, Figure I shows that they are concentrated in neighborhoods with higher poverty rates: 58% of evictions in New York and 46% of evictions in Cook County occur in high-poverty neighborhoods, which are defined as census tracts with more than 20% of residents living below the poverty line. This spatial concentration is consistent with Desmond (2012), Desmond and Kimbro (2015), and Desmond and Gershenson (2017), who find that eviction is common in poor communities in Milwaukee. Online Appendix Figure B.1 shows how eviction filing rates (the number of evictions filed relative to the number of occupied rental units) vary across neighborhoods. Some neighborhoods have annual eviction filing rates as high as 1 in 10 renter households in Cook County and as high as 1 in 5 renter households in New York.

Figure I.

Evictions and Neighborhood Poverty

This figure depicts the approximate locations of court-ordered evictions in Cook County (left) and New York (right) in 2010 (each dot represents 20 eviction orders in the census tract), along with the poverty rate of the census tract (based on 2006–2010 American Community Survey five-year averages).

While evictions primarily occur in neighborhoods with high poverty rates, tenants in our linked housing court sample are also negatively selected on precourt earnings and employment relative to randomly chosen tenants who live in the same neighborhoods. Table I shows descriptive statistics for three groups: evicted tenants, nonevicted tenants with a case filed in housing court, and ACS respondents who are renters, weighted so the distribution of their neighborhoods matches the distribution of neighborhoods for tenants in our sample of eviction cases. Relative to renters from the same neighborhoods, tenants in the linked sample have lower levels of earnings and employment than renters from the same neighborhoods. Within housing court, differences persist, with evicted tenants showing lower levels of earnings and employment than nonevicted tenants. For example, in Cook County, average quarterly earnings in the eight quarters before case filing are 4,876 for nonevicted tenants and 3,907 for evicted tenants; in New York these numbers are 3,628 and 3,080, respectively.³¹

TABLE I.

Linked Sample Summary Statistics

	Cook County			New York
			Renters			Renters
	Evicted	Not	from same	Evicted	Not	from same
		evicted	neighbor-		evicted	neighbor-
			hoods			hoods
	(1)	(2)	(3)	(4)	(5)	(6)
Individual characteristics
Age	37.51	37.44	33.98	38.55	40.49	34.98
	(10.35)	(10.22)	(10.26)	(9.50)	(9.22)	(10.51)
Female	0.62	0.62	0.54	0.71	0.74	0.54
	(0.49)	(0.49)	(0.50)	(0.46)	(0.44)	(0.50)
Black	0.69	0.66	0.47	0.58	0.58	0.35
	(0.46)	(0.47)	(0.46)	(0.49)	(0.49)	(0.43)
Hispanic	0.12	0.11	0.20	0.46	0.47	0.39
	(0.32)	(0.31)	(0.42)	(0.50)	(0.50)	(0.47)
Quarterly earnings	3,907	4,876	6,237	3,080	3,628	7,059
	(4,636)	(5,561)	(9,251)	(4,066)	(4,471)	(14,987)
Employment	0.58	0.61	0.72	0.47	0.51	0.70
	(0.46)	(0.47)	(0.43)	(0.42)	(0.44)	(0.45)
Neighborhood poverty rate (5 yr. avg.)	0.21	0.20	0.20	0.29	0.29	0.24
	(0.13)	(0.13)	(0.09)	(0.12)	(0.12)	(0.10)
Neighborhood median rent (5 yr. avg.)	762	788	824	943	933	950
	(195)	(229)	(172)	(208)	(226)	(298)
Case characteristics
Ad damnum (1,000s)	2.01	1.74		4.60	4.16
	(2.99)	(3.09)		(27.79)	(31.45)
No prior case	0.63	0.67		0.54	0.53
	(0.48)	(0.47)		(0.50)	(0.50)
Tenant without attorney	0.97	0.94		1.00	0.99
	(0.16)	(0.23)		(0.07)	(0.10)
Observations	193,000	108,000	36,559	87,294	70,474	103,614

Notes. The statistics in columns (1), (2), (4), and (5) are for the samples matched to earnings and employment records. Columns (1) and (4) include summary statistics for those in housing court who are evicted. Columns (2) and (5) include summary statistics for those in housing court who are not evicted. For these samples, quarterly earnings is the average quarterly earnings in quarters 1–8 before filing, and employment is the fraction of quarters with positive earnings in quarters 1–8 before filing. Columns (3) and (6) include summary statistics for renters aged 18–55 in the ACS PUMS 2006–2010, weighted to match the distribution of neighborhoods (Public Use Microdata Areas) for tenants who have housing court cases filed against them. For the ACS samples, quarterly earnings is the annual wage income divided by four, and employment is approximated by the proportion of people with any wage income. For the linked samples, neighborhood poverty rate and neighborhood median rent are based on census tract. For the ACS sample, these outcomes are based on Public Use Microdata Areas. See Online Appendix Table F.1 for a similar table on financial health outcomes for cases linked to credit bureau records. Cook County observation counts are rounded in accordance with Census Bureau disclosure requirements. Cook County results were approved for release by the U.S. Census Bureau, authorization no. CBDRB-FY22-072.

Table I further shows that relative to renters from the same neighborhoods, both evicted and nonevicted tenants in our linked sample are more likely to be female (62% versus 54% in Cook County and 71% versus 54% in New York) and more likely to be Black (68% versus 47% in Cook County and 58% versus 35% in New York). Hispanic tenants are underrepresented in our linked housing court sample in Cook County (12% versus 20%), but overrepresented in NYC (46% versus 39%). By contrast, the demographic characteristics of evicted and nonevicted tenants within housing court are similar.

The bottom panel of Table I displays case characteristics. The average ad damnum amount—the judgment amount the landlord is seeking from the court—for evicted tenants is around 2,000 in Cook County and 4,600 in New York, both of which are a few hundred dollars more than for nonevicted tenants. In Cook County, evicted tenants are less likely than nonevicted tenants to have no prior case (63% versus 67%) and somewhat more likely to be unrepresented (97% versus 94%), while in New York, evicted and nonevicted tenants are somewhat more similar in these respects (53% versus 54% have no prior case and nearly all tenants in NYC are unrepresented at the time of the initial hearing).³²

IV.B. Trends around Court Filing

We plot the trends in our main outcomes for evicted and nonevicted tenants relative to the time the eviction case is filed. Figure II shows these trends for earnings, employment, residential mobility, neighborhood poverty, emergency shelter use, and use of homelessness services; Figure III shows these trends for financial outcomes; Figure IV shows these trends for health outcomes.

Figure II.

Labor Market and Housing Outcomes Relative to the Time of Eviction Filing

This figure shows trends in labor market and housing outcomes relative to eviction filing, combined across Cook County and New York. For each location, we estimate equation (1) and plot the equal-weighted average for the evicted and nonevicted groups in each time period. The only controls are calendar year dummies. For both sets of coefficients, we add in the nonevicted group mean in the omitted period so that the magnitudes are easy to interpret. The employment and earnings outcomes are measured at a quarterly frequency, while the housing outcomes are measured at an annual frequency. Online Appendix E shows these trends by location. Cook County results were approved for release by the U.S. Census Bureau, authorization no. CBDRB-FY22-072.

Figure III.

Financial Health Outcomes Relative to the Time of Eviction Filing

This figure plots trends in credit report outcomes relative to eviction filing, combined across Cook County and New York. For each location, we estimate equation (1) and plot the equal-weighted average for the evicted and nonevicted groups in each time period. The only controls are calendar year dummies. For both sets of coefficients, we add in the nonevicted group mean in the omitted period so that the magnitudes are easy to interpret. All outcomes are measured at a quarterly frequency. Online Appendix E shows results by location.

Figure IV.

Health Outcomes Relative to the Time of Eviction Filing (New York)

This figure shows trends in health outcomes relative to eviction filing in New York. We observe health outcomes in the New York sample only. We estimate equation (1) and plot results for the evicted and nonevicted groups in each time period. The only controls are calendar year dummies. For both sets of coefficients, we add in the nonevicted group mean in the omitted period so that the magnitudes are easy to interpret. All outcomes are measured at a quarterly frequency.

The figures are based on the regression

(1)

where r indexes time relative to the eviction filing, E_i is an indicator for the case ending in an eviction order, β_r are coefficients on indicators for time relative to the case filing, and δ_r are coefficients on indicators for relative time interacted with the eviction outcome. The only controls we include are calendar year dummies γ_t. The start and end periods are S and F, respectively, and O is the omitted period. We estimate equation (1) separately by location and present equal-weighted averages. Location-specific trends are presented in Online Appendix E. Figures II–IV display regression estimates of β_r and α + δ_r + β_r, with the nonevicted group mean in the omitted period added to both sets of coefficients so that the magnitudes are easy to interpret. Since we add the mean in the omitted period, the levels in the figure are not sensitive to the choice of omitted period.

The top two panels of Figure II depict trends in quarterly earnings and employment—the result of estimating equation (1) between 16 quarters prior to filing and 24 quarters after filing. Both evicted and nonevicted groups show signs of declining earnings in the year before case filing. This decline is steeper for tenants who are evicted (−340) than for those who are not evicted (−157). Similarly, the probability of being employed for evicted and nonevicted tenants declines in the year prior to filing, with the decline for evicted tenants more severe (−1.3 percentage points) than for nonevicted tenants (−0.5 percentage points). Following eviction, employment does not recover to its prefiling peak over the next six years. There is a slight tapering of employment for the entire sample period after filing, which is not due to aging into retirement since our sample includes individuals between 18 and 55 years old at the time of the eviction filing.

Turning next to residential mobility, Figure II, Panel C shows the probability that we observe a tenant at an address different from the filing address. We study the same time window as for employment and earnings, now at the annual frequency that is imposed by the MAFARF. In the year of filing, 22% of tenants are observed at an address different from that recorded in the case. The fact that this estimate is not zero reflects moves in the year of filing and noise in the mobility data. The probability of observing a tenant at a new address increases to about 37 percent for the evicted group in the first year after filing and rises to 81% six years after filing. This probability rises faster for evicted than for nonevicted tenants, yielding a gap of about 16 percentage points six years after filing.³³ This gap may be an underestimate if evicted individuals are less likely to have updated addresses, which we find some evidence of in Online Appendix Figure E.1.³⁴ Although evictions are associated with increased residential mobility, Panel D shows that there is little change in the average neighborhood poverty rate before or after the case is filed.

One of the most striking results is that the use of homelessness services spikes in the year after filing, particularly for the evicted group (Figure II, Panel E).³⁵ The relative magnitudes of these increases are sizable: for the evicted group, the probability of using homelessness services increases from 1.4% in the filing year to 7.1% in the first year after filing, an increase of approximately 400%. The nonevicted group also increases their use of homelessness services over the same period but the increase is smaller, from 1.3% to 1.9%. Panel F shows that this increase in use of homelessness services is primarily due to increased use of emergency shelters: for evicted tenants, the probability of using an emergency shelter jumps from 1% to 6% between the year of case filing and the following year.

Next we examine trends in financial health, presented in Figure III. We study trends between eight quarters prior and 20 quarters after the case filing because there are fewer years available in the credit bureau sample in New York. Mirroring the trends in earnings, the financial health index declines in the year prior to filing by roughly 0.067 std. dev. for nonevicted tenants and 0.085 std. dev. for evicted tenants. Looking at the index’s components, credit scores fall, unpaid bills rise, and access to credit decreases in the year before filing.³⁶ These figures reveal that tenants facing eviction are financially distressed prior to court: they have low average credit scores and high levels of indebtedness in the years prior to housing court, and the mean tenant would be considered a subprime borrower. Following the eviction case, tenants have diminished financial health—including elevated indebtedness and diminished credit access—for several years regardless of the outcome of the court case. In the four years after the case, the financial health index does not return to its prefiling peak for either group. The gap in financial health between evicted and nonevicted tenants also widens in the aftermath of eviction court, increasing from about −0.14 std. dev. two years prior to the case to about −0.18 std. dev. two quarters after the case (before tapering slightly over the next two years). While the gap in unpaid bills that arises immediately following the case closes by quarter 4, the gap in access to credit widens in the aftermath of the court case. The difference in the likelihood of having no source of revolving credit is about 3.7 percentage points four quarters before the case, rises to about 5.4 percentage points by one quarter after the case, and remains elevated through quarter 12.

Figure IV shows trends in total hospital visits, total emergency room visits, and total hospital visits related to mental health in the New York sample. Panel A shows that total hospital visits increase in the two years leading up to the eviction filing and peak during the quarter of filing, which coincides with the point where earnings are at their lowest. The increase preceding housing court hints at the possibility that health shocks could be a source of earnings losses that lead to nonpayment of rent, although it is not clear in which direction causality runs. Panel B shows that the vast majority of these hospital visits are trips to the emergency room, and Panel C shows that the total number of mental health–related hospital visits also increases during the period leading up to housing court. The gap between evicted and nonevicted tenants in hospital visits widens following eviction in all three panels.

IV.C. Considerations for Empirical Analysis

The analysis up to this point has revealed patterns that are consistent with changes to prefiling earnings, health, and financial circumstances being correlated with both the case filing and with receiving an eviction order. Evicted tenants have lower earnings, worse credit, and higher rates of hospitalization than nonevicted tenants several years before filing, and they experience sharper drops in earnings, and steeper jumps in unpaid bills and hospital visits in the immediate run-up to filing. This raises concerns about selection on correlated unobservables at both the filing and the eviction stage.

The presence of such correlated unobservables can bias frequently used methods for identifying the effects of eviction, such as cross-sectional comparisons corrected only for observable characteristics, and difference-in-differences methods. We explore the potential bias of such methods using our data. We first examine what a simple demographic- and location-adjusted cross-sectional comparison of evicted tenants to renters outside of court would yield for the effects of eviction on earnings. The result appears as the leftmost bar in Online Appendix Figure F.1 and implies that eviction reduces average quarterly earnings by roughly 1,600 in Cook County and 1,100 in New York. Moving to a within-court comparison of evicted and not evicted tenants (middle bar), the estimates shrink by approximately one third to 1,000 in Cook County and 600 in New York. This suggests that comparisons of tenants outside of court to those inside will likely overstate the effect of eviction because they will incorrectly attribute selection into court to the eviction itself.³⁷

While the estimates shown in the second bar remove bias due to selection into court, they do not address bias stemming from the selection on levels or trends within court. Difference-in-differences (DiD) is a natural choice of method for addressing selection on levels. The third bar of Online Appendix Figure F.1 shows estimates from a DiD specification.³⁸ Adjusting for differences in levels between evicted and not evicted in the lead-up to case filing shrinks the estimates further. However, the differential pretrends among evicted and nonevicted tenants in Figures II, III, and IV still raise concerns about bias in DiD estimates. Although it is possible to outline assumptions under which bias is not a concern or under which the bias can be signed (see Heckman and Robb 1985), we instead rely on our quasi-experimental instrumental variables research design, which we describe in the next section. This design addresses the sources of selection that we document above and allows us to identify a local average treatment effect of eviction.

V. Empirical Framework

This section describes our IV approach based on judges’ tendency to evict in cases randomly assigned to them. We discuss how the assumptions that underlie this identification strategy are supported by the institutional environment and provide tests of these assumptions. We also describe how we combine estimates across locations.

V.A. Instrumental Variables

The evidence in Section IV suggests that whether a tenant is evicted may depend on unobserved characteristics and unobserved shocks that affect both eviction and subsequent outcomes. If a suitable instrument is available, it can be used to solve this endogeneity problem and estimate causal effects of eviction. A common approach in court settings is to exploit the random assignment of cases to judges and use Z_j(i) as an IV, where Z_j(i) is the leave-one-out estimate of stringency for judge j assigned to individual i’s case. This approach estimates the following two-stage least squares model:

(2)

(3)

where the least squares regression is run separately for each outcome and time period. Here E_i is an indicator for whether case-individual i has an eviction, Y_i is the observed outcome, and X_i is a set of controls for individual and case characteristics. Controls include court-by-year-quarter fixed effects, ad damnum amount, gender, race indicators, census tract poverty rates, census tract rent, a cubic in age at filing date, and indicators for missing controls.³⁹ Our main OLS and IV specifications include additional controls for lagged values of the dependent variable, which are described in the table notes. If the IV assumptions are satisfied, this analysis will recover a positive weighted average effect of eviction among compliers, where compliers are defined as tenants who would have received a different eviction outcome had their case been assigned to a different judge (Imbens and Angrist 1994).

1. The Judge Stringency Instrument

We measure judge stringency using the yearly leave-one-out mean eviction rate for the initial judge (Cook County) or courtroom (New York) assignment. Using the sample described in Section III, we calculate the stringency of the judge to which tenant i’s case is assigned, Z_j(i), as the leave-one-out mean eviction rate (omitting i) for judge j(i) in the same year. In a typical year, there are 21 judges in Cook County and 29 courtrooms in New York hearing cases. Over our sample period, we observe 127 unique judges in Cook County. We construct the instrument from an average of 1,600 cases per judge (per year) in Cook County and 3,400 per courtroom (per year) in New York.

Figure V shows the distribution of judge stringency, residualized by court-year-quarter, across cases in Cook County and New York. The variation in judge stringency is substantial and similar across locations: a 7 percentage point difference between the 10th percentile and 90th percentile of judge stringency in Cook County and a 6 percentage point difference in New York.

Figure V.

Judge Stringency

For each location, this figure shows a histogram of judge stringency, residualized by court-year-quarter, with the number of cases indicated along the left vertical axis. Each panel also depicts fitted values from a local linear first-stage regression of eviction on judge stringency and court-year-quarter fixed effects (solid line, plotted along the right vertical axis). Dotted lines show 95% confidence intervals.

2. Validating the IV Design

This section discusses conditions for judge stringency to be a valid instrument and for the IV estimand to be interpretable as a positive weighted average of local treatment effects on compliers: relevance, exogeneity, exclusion, and monotonicity. We discuss each assumption and support them with arguments based on institutional details and empirical evidence.

1. Relevance.  Table II, columns (1) and (3) report first-stage estimates from equation (2) for Cook County and New York, respectively. Judge stringency has a large and statistically significant impact on evictions, with a partial F-statistic for judge stringency of 934 in Cook County and 288 in New York, relieving concerns about weak instruments. Columns (2) and (4) show that the first-stage coefficients change very little when we include controls, consistent with judge stringency being uncorrelated with individual and case characteristics. Online Appendix Table G.2 additionally reports F-statistics for the Black and female subgroups that we study in Section VI.

   Online Appendix G.A provides additional robustness checks on the first stage. We show that the first stage is robust to (i) controlling for other dimensions of judge behavior, (ii) using an alternate approach to measuring the first judge or courtroom in the court records, and (iii) using different sample selection criteria.

2. Exogeneity.  Table III shows the result of a standard balance test of random assignment. As we would expect, columns (1) and (3) show that case and tenant characteristics predict receiving an eviction order in both locations. Importantly, columns (2) and (4) show that these characteristics are not predictive of the stringency of the judge randomly assigned to the case. Only one of the 17 coefficients in columns (2) and (4) is statistically significant and is quantitatively small (a −0.017 std. dev. decrease in stringency for joint action cases). In addition, we fail to reject the null hypothesis that the coefficients are jointly equal to zero in both locations, consistent with random assignment. Online Appendix G.B provides additional evidence that judge stringency is uncorrelated with lagged values of our key outcomes.

3. Exclusion. Our estimation strategy relies on the assumption that judge stringency affects tenant outcomes only through the eviction order. As discussed in Section II, judges determine whether to issue an eviction order but may also influence other aspects of the case, such as the judgment amount (in cases in which the landlord is seeking rental arrears or damages) or whether a stay of enforcement is granted (which allows extra time for the tenant to move before an enforcement). The multidimensionality of judge discretion could make it challenging to isolate the impact of the eviction order (Mueller-Smith 2015; Bhuller et al. 2020).

   Exclusion will be violated if judge stringency is correlated with other dimensions of judge discretion that affect tenant outcomes. To assess the plausibility of the exclusion restriction, we first examine whether eviction order stringency is correlated with other dimensions of judge stringency. Online Appendix Tables G.6 and G.8 report pairwise correlations between eviction order stringency (the instrument) and stringency constructed along alternative dimensions of the case.⁴⁰ In each instance, the correlations are weak. Next, in Online Appendix Table G.7 we reestimate our first stage with and without these alternative stringency measures and find that including these measures has minimal effect on the first-stage coefficient, providing additional support for the plausibility of the exclusion restriction. In addition, in Cook County—where we can observe the judgment amount—we reestimate the main IV regressions for housing, labor, and financial outcomes in the first year, with an additional control for judgment amount stringency, and find that the main conclusions are unchanged.

   Finally, the practical aspects of case proceedings provide additional reassurance that judge discretion in judgment amounts is unlikely to be a threat to our research design. For instance, we find the judgment amount for a case is closely linked to the amount the landlord initially requests in the filing. In Cook County, the correlation between the judgment amount and the ad damnum amount is 0.81. This lends support for the idea that judges’ differences along this dimension are likely to be small and unlikely to be driving our results. Taken together, the robustness checks in Online Appendix G.C suggest that the multidimensionality of judge discretion is unlikely to be a threat to the exclusion restriction in our settings.

4. Monotonicity. For the IV estimates to be interpreted as a positively weighted average of local average treatment effects (LATEs), we need monotonicity to be satisfied (Imbens and Angrist 1994). In our setting, monotonicity requires that evicted tenants would also have been evicted by a more stringent judge, while nonevicted tenants would not have been evicted by a less stringent judge. This condition can fail in randomized judge designs if judges are relatively harsh for some types of cases or individuals and relatively lenient for others, or if judges differ in both diagnostic skills and preferences regarding the outcome of the case, as discussed by Chan, Gentzkow, and Yu (2022). We perform two tests of this assumption. First, under monotonicity, the first-stage estimates should be nonnegative for any subsample of tenants. Online Appendix Tables G.11 and G.12 show nonnegative first-stage estimates for various subsamples in Cook County and New York. As a second test, we calculate judge stringency on one subpopulation (for example, women) and then use that stringency measure in the first stage for the complementing subpopulation (for example, men), as in Bhuller et al. (2020) and Norris, Pecenco, and Weaver (2021). Online Appendix Table G.13 presents this test and shows nonnegative and similar-sized first-stage estimates across specifications. Hence, neither of these tests provide evidence against the monotonicity assumption.

TABLE II.

First Stage

	Cook County		New York
	(1)	(2)	(3)	(4)
Judge stringency	0.741***	0.740***	0.831***	0.825***
	(0.025)	(0.024)	(0.057)	(0.057)
Controls	No	Yes	No	Yes
Observations	268,000	268,000	150,662	150,662

Notes. This table reports results from the first-stage regression of eviction on judge stringency, for Cook County and New York using the linked labor market sample. Columns (1) and (3) include our judge stringency measure with court-year-quarter fixed effects, but without individual controls. Columns (2) and (4) add controls. The additional controls include ad damnum amount, gender, race indicators, census tract poverty rate, census tract rent, a cubic in age at filing date, and indicators for missing controls. Online Appendix Table G.1 provides additional evidence on the robustness of the first-stage regression. Cook County observation counts are rounded in accordance with Census Bureau disclosure requirements. Cook County results were approved for release by the U.S. Census Bureau, authorization no. CBDRB-FY22-072. * p < .1, ** p < .05, *** p < .01.

V.B. Combining Estimates across Locations

Due to restrictions on data sharing, we are unable to pool individual observations from Cook County and New York. We therefore estimate each specification separately by location and then report average point estimates in the tables in Section VI, along with each location-specific estimate. The combined point estimates weight results from the two locations equally, and we calculate the standard errors for the combined estimates as

where ω = 0.5. Under the assumptions outlined in Section V.A, the combined estimates can be interpreted as the average effect of eviction for compliers in Cook County and New York.

VI. Results

This section presents OLS and IV estimates of the effects of eviction on tenants’ residential mobility, use of homelessness services, labor market outcomes, financial strain, and hospital use. The estimates show that eviction increases residential mobility and causes spikes in emergency shelter use and hospital visits (particularly for mental health–related conditions) in the year after filing. Housing instability persists in the second year after filing, with eviction triggering increased use of homelessness services. These findings suggest a period of instability of at least two years. During this period, evicted tenants also experience reductions in earnings. In the longer run, we find that eviction worsens financial health through increased indebtedness and reductions in credit scores.

VI.A. Eviction Order Enforcement, Residential Mobility, and Neighborhood Poverty

We first study how eviction affects a tenant’s housing situation, focusing on enforced eviction orders, residential moves, and neighborhood poverty. We report estimates for the full sample, separately by location, and separately for female and Black tenants. We focus on female and Black tenants in our subgroup analysis because these groups are overrepresented in housing court in Cook County and New York and because prior research suggests they may face greater adverse consequences of eviction. Qualitative research (Desmond et al. 2013; Desmond 2016) points to two potential reasons for more severe effects of eviction on women, both revolving around children in the household. First, as a result of both greater childcare responsibilities and larger household size, women may face more difficulties securing and maintaining new accommodation (Sugrue 2005; Desmond 2012; U.S. Department of Housing and Urban Development 2018). Second, landlords may be reluctant to rent to households with children because children may cause nuisances to neighbors or attract inspections by Child Protective Services or the city’s health department for lead hazards (Roberts 2001). Black households may experience more adverse impacts of eviction because of discrimination while searching for new housing (Bayer et al. 2017; Christensen and Timmins 2022), which would exacerbate the disruptive effects of eviction (Desmond and Gershenson 2016).

1. Order Enforcement

To better characterize the treatment, we first consider the extent to which eviction orders are enforced by a sheriff or marshal. This experience may cause tenants to move out more quickly or unexpectedly, leaving them unable to secure new housing before they are locked out. In addition to potentially increasing the likelihood of moving, an eviction order may change the circumstances under which households move. Table IV shows that receiving an eviction order substantially increases the probability of experiencing an enforcement within one year, with an IV estimate of 43.5 percentage points and an OLS estimate of 30.1 percentage points. Moves occurring after enforced orders may be more likely to occur under greater stress and exigency and may potentially result in moves to lower-quality neighborhoods or homelessness. We investigate the effects on neighborhood quality and homelessness below.

TABLE IV.

Impact on Housing Situation

	1 year after filing			2 years after filing
		OLS	IV		OLS	IV
	(1)	(2)	(3)	(4)	(5)	(6)
Enforcement	0.002	0.301***	0.435***	0.002	0.313***	0.422***
	(0.031)	(0.005)	(0.039)	(0.032)	(0.005)	(0.037)
	[329,279]			[307,837]
By location
Cook County	0.004	0.270***	0.451***	0.004	0.275***	0.442***
	(0.059)	(0.004)	(0.050)	(0.062)	(0.004)	(0.049)
New York	0.000	0.333***	0.419***	0.000	0.351***	0.401***
	(0.017)	(0.009)	(0.060)	(0.018)	(0.010)	(0.057)
By group
Female	0.002	0.290***	0.425***	0.002	0.302***	0.418***
	(0.030)	(0.005)	(0.046)	(0.032)	(0.005)	(0.045)
Black	0.002	0.307***	0.464***	0.002	0.319***	0.436***
	(0.032)	(0.005)	(0.046)	(0.035)	(0.006)	(0.044)
Not at eviction address	0.293	0.073***	0.082**	0.478	0.129***	0.111**
	(0.318)	(0.003)	(0.036)	(0.348)	(0.003)	(0.053)
	[218,228]			[183,227]
By location
Cook County	0.363	0.031***	0.093	0.568	0.070***	0.074
	(0.481)	(0.003)	(0.057)	(0.495)	(0.004)	(0.064)
New York	0.222	0.116***	0.071	0.389	0.188***	0.149*
	(0.415)	(0.006)	(0.045)	(0.487)	(0.004)	(0.084)
By group
Female	0.280	0.081***	0.093**	0.461	0.139***	0.136**
	(0.312)	(0.004)	(0.046)	(0.343)	(0.003)	(0.060)
Black	0.272	0.079***	0.066	0.454	0.138***	0.098
	(0.310)	(0.004)	(0.056)	(0.345)	(0.004)	(0.080)
Neighborhood poverty rate	0.247	−0.000	−0.002	0.246	−0.001	−0.008
	(0.088)	(0.000)	(0.010)	(0.090)	(0.001)	(0.014)
	[173,909]			[127,891]
By location
Cook County	0.195	0.001	−0.014	0.196	0.002**	0.011
	(0.130)	(0.001)	(0.018)	(0.133)	(0.001)	(0.021)
New York	0.298	−0.001***	0.009	0.295	−0.004***	−0.027
	(0.120)	(0.000)	(0.007)	(0.123)	(0.001)	(0.020)
By group
Female	0.258	−0.000	−0.004	0.256	−0.001	−0.009
	(0.090)	(0.001)	(0.012)	(0.091)	(0.001)	(0.020)
Black	0.267	−0.001	−0.005	0.266	−0.003***	−0.023
	(0.088)	(0.001)	(0.013)	(0.090)	(0.001)	(0.022)

Notes. This table reports equally weighted averages of Cook County and New York nonevicted sample means , as well as equally weighted averages of location-specific OLS (OLS) and two-stage least squares (IV) estimates of the effect of eviction on outcomes related to the tenant’s housing situation. Outcomes are listed on the left of each row. Results are shown for one year (columns (1)–(3)) and for two years (columns (4)–(6)) after an eviction case is filed. Each panel shows results for a given outcome. Below the combined estimates in each panel, we report estimates separately for each location and for the female and Black subsamples. “Enforcement” is defined as an enforcement officer executing the eviction order associated with the case, and is defined cumulatively. “Not at eviction address” is an indicator for being observed living at a different address than the filing address. “Neighborhood poverty rate” is the tract-level poverty rate. The main set of controls included in all model specifications are ad damnum amount, gender, race, census tract poverty rates, census tract rent, a cubic in age at filing date, dummies for missing controls, and court-by-year fixed effects. In Cook County regressions only, we also include an indicator for case type. For “Not at eviction address” and “Neighborhood poverty rate” we also control for whether tenants were not at the eviction address one year and two years prior and tenants’ neighborhood poverty rate one year and two years prior to the eviction case filing, respectively. Standard errors for regression model coefficients are included in parentheses and are clustered at the judge(courtroom)-year level. Observation counts for the full sample are listed under the standard errors, in brackets. Observation counts for all outcomes and subgroups can be found in Online Appendix Table H.15. The reduced-form results for all outcomes can be found in Online Appendix Table H.10. Cook County results were approved for release by the U.S. Census Bureau, authorization no. CBDRB-FY22-072. * p < .1, ** p < .05, *** p < .01.

2. Residential Mobility

As we showed in Section IV, tenants in housing court have high move rates regardless of the case outcome, with evicted tenants being more mobile both before and after the case. The IV models allow us to estimate how much additional residential mobility is caused by an eviction. Table IV shows that for compliers, receiving an eviction order increases the probability of appearing at a new address by 8.2 percentage points one year after filing (an increase of 28% relative to a mobility rate of 29.2% for the nonevicted group). The OLS estimate is similar though slightly smaller (7.3 percentage points). In Online Appendix I, we explore alternative approaches to defining moves and find that eviction increases residential mobility under a variety of alternative definitions.⁴¹ The effects of eviction on residential mobility are similar across locations and subgroups. Columns (4)–(6) show that these effects persist into the second year, and the IV estimate implies that eviction increases the probability of not being at the eviction address by 11.1 percentage points (23%).⁴² In both time periods, the effects on residential mobility are larger for women.⁴³ After year 2, the effect of eviction on moving residences diminishes and becomes statistically insignificant (see Online Appendix H.B), although these effects are estimated with less precision. This result is consistent with the interpretation that after year 2, the causal effect of the eviction order on moving residences is muted by the nonevicted group becoming more likely to move.

3. Neighborhood Quality

In the bottom panel of Table IV, we consider the effect of an eviction order on neighborhood quality, as measured by the census tract poverty rate. We find little evidence that eviction causes tenants to move to neighborhoods with higher poverty rates, either in the combined estimates, the location-specific estimates, or the demographic-specific estimates. These estimates are fairly precise, and we can rule out an effect on the neighborhood poverty rate of more than 2.2 percentage points for the combined sample with 95% confidence. Neither our IV nor our OLS specifications point to an increase in the neighborhood poverty rate. Individuals at risk of eviction live in high-poverty neighborhoods prior to filing, which may help explain why eviction does not cause them to move to even higher-poverty neighborhoods on average. Our findings contrast with Desmond and Shollenberger (2015), who find that among recent movers, those who experience a forced move relocate to neighborhoods with 5 percentage points higher poverty rates.⁴⁴ Given that tenants who are evicted move to observably similar neighborhoods, the effects we consider below on other socioeconomic outcomes likely do not arise due to changes in neighborhood environment, as in studies of housing mobility programs (Chetty, Hendren, and Katz 2016) or public housing demolitions (Chyn 2018).

VI.B. Homelessness

Homelessness carries substantial private and social costs (Evans, Philips, and Ruffini 2019). Although eviction has the potential to be a direct cause of homelessness, there is currently no causal evidence on this relationship. The event studies in Section IV show a striking increase in homelessness after filing for evicted tenants, which suggests a causal link. We investigate this link directly using our IV research design. In addition to their policy relevance, the effects on homelessness are informative as a measure of material hardship and a possible mechanism for the labor market effects studied in Section VI.C.

Table V shows that an eviction order increases the probability of using emergency shelter in the year after filing by 3.4 percentage points in the IV specification and 3.1 percentage points in the OLS specification, which are both large relative to the nonevicted mean of 0.9%. We don’t find evidence of increased use of emergency shelters after the first year, as seen in column (6) and in the longer-run results presented in Online Appendix Table H.5. Similarly, the OLS estimates are approximately half as large after the first year. These results suggest that evicted tenants experience difficulty finding alternative housing in the immediate aftermath of the court case and are consistent with economic models of homelessness that emphasize the transitory dynamics of homelessness (O’Flaherty 2004).

TABLE V.

Effect on the Use of Homelessness Shelter and Services

	One year after filing				Two years after filing
		OLS	IV			OLS	IV
	(1)	(2)	(3)		(4)	(5)	(6)
Emergency shelter	0.009	0.031***	0.034**		0.008	0.014***	−0.001
	(0.068)	(0.001)	(0.017)		(0.062)	(0.001)	(0.013)
	[210,840]				[198,898]
By location
Cook County	0.007	0.010***	0.023		0.006	0.006***	−0.019
	(0.086)	(0.002)	(0.028)		(0.077)	(0.001)	(0.019)
New York	0.011	0.052***	0.046**		0.009	0.022***	0.016
	(0.105)	(0.002)	(0.019)		(0.097)	(0.001)	(0.017)
By group
Female	0.009	0.032***	0.024		0.008	0.016***	0.024
	(0.066)	(0.002)	(0.018)		(0.061)	(0.001)	(0.015)
Black	0.010	0.034***	0.036		0.009	0.015***	0.007
	(0.072)	(0.002)	(0.024)		(0.068)	(0.001)	(0.019)
Any homelessness services	0.015	0.036***	0.029		0.012	0.019***	0.036**
	(0.086)	(0.002)	(0.023)		(0.076)	(0.001)	(0.015)
	[210,840]				[198,898]
By location
Cook County	0.017	0.016***	0.012		0.012	0.013***	0.048**
	(0.128)	(0.002)	(0.042)		(0.110)	(0.002)	(0.023)
New York	0.013	0.056***	0.046**		0.011	0.025***	0.024
	(0.114)	(0.002)	(0.018)		(0.104)	(0.001)	(0.019)
By group
Female	0.015	0.038***	0.030		0.012	0.020***	0.068***
	(0.086)	(0.002)	(0.023)		(0.077)	(0.001)	(0.020)
Black	0.017	0.040***	0.049*		0.014	0.020***	0.057***
	(0.092)	(0.002)	(0.029)		(0.083)	(0.001)	(0.022)

Notes. This table reports equally weighted averages of Cook County and New York nonevicted sample means , as well as equally weighted averages of location-specific OLS (OLS) and two-stage least squares (IV) estimates of the impact of eviction on outcomes related to the tenant’s homelessness situation. Outcomes are listed on the left of each row. Results are shown for one year (columns (1)–(3)) and two years (columns (4)–(6)) after the eviction case is filed. Each panel shows results for a given outcome. Below the combined estimates in each panel, we report estimates separately for each location and for the female and Black subsamples. “Emergency shelter” is an indicator for if the individual was observed staying at an emergency homeless shelter. “Any homelessness services” is an indicator for having any interaction with homelessness services. The main controls for all model specifications are the same as those described in Table IV. In each regression, we control for whether tenants stayed at an emergency shelter and whether tenants had any interaction with emergency homelessness services one year and two years prior to the eviction case filing. Standard errors for regression model coefficients are included in parentheses and are clustered at the judge(courtroom)-year level. Observation counts for the main combined specifications are reported in brackets below the standard errors in columns (3) and (6). Observation counts for all regressions shown here can be found in Online Appendix Table H.15. The reduced-form results for regressions shown above can be found in Online Appendix Table H.11. Cook County results were approved for release by the U.S. Census Bureau, authorization no. CBDRB-FY22-072. * p < .1, ** p < .05, *** p < .01.

We find similar effects in the first year when looking at use of any homelessness service for both IV and OLS, though the IV estimate is not statistically significant. While the effects on shelter use are concentrated in the year after filing, the effect on using any homelessness service remains elevated beyond the first year. The IV estimates indicate that evicted tenants are 3.6 percentage points more likely to use homelessness services than tenants who avoid eviction in the second year after filing (an increase of 200% relative to the nonevicted group mean of 1.2 percentage points). As with residential mobility, longer-term interactions with homelessness services are driven by effects for female and Black tenants, with an IV estimate for female tenants of 6.8 percentage points (467%) and an IV estimate for Black tenants of 5.7 percentage points (307%).

The results above indicate that eviction causes a large increase in homelessness in the first year after a case (through increases in emergency shelter use) and beyond (through elevated use of homelessness services). We view these results as complementary to work on short-term emergency financial assistance and homelessness (Evans, Sullivan, and Wallskog 2016), which finds that temporary assistance to at-risk tenants can lead to persistent reductions in homelessness. These results also connect to research emphasizing the socioeconomic consequences of changes to proceedings in eviction court (Greiner, Pattanayak, and Hennessy 2012). While homelessness remains rare, even for tenants in eviction court, our estimates nevertheless imply substantial additional homelessness caused by evictions. In a given year, across our two locations, we estimate that evictions lead to more than 3,600 adults staying in emergency shelter in the year after filing and 2,500 adults using homelessness services the following year. These estimates are likely to understate the overall effect on homelessness for two reasons: we only estimate the effects of eviction for individuals named on a lease, which will leave out other household members, such as children, and our measures of homelessness will miss effects on unsheltered spells of homelessness.

VI.C. Earnings and Employment

We now shift attention to estimates of the causal effects of an eviction order on earnings and employment. Table VI reports estimates for quarters 1–4 and 5–8 after case filing. The first row reports the combined estimates for earnings. The IV estimate shows that eviction decreases average quarterly earnings in quarters 1–4 by 323 (7% of the nonevicted mean of 4,300). This effect is similar in magnitude to the earnings drop among evicted tenants in the year prior to filing. The effects of eviction on earnings are larger in the second year after filing, reducing average quarterly earnings by 613 (14% of the nonevicted mean). The point estimates are larger for female and Black tenants, although formal tests of equality fail to reject a null hypothesis of equality (see Online Appendix Table H.1). The estimated effects are also comparable across the two locations.⁴⁵ Comparing the IV and OLS estimates, the OLS estimates are systematically smaller, suggesting that effects may be larger for compliers. As discussed in Section III, in New York, earnings are not observed when an individual moves out of state, and in Cook County, earnings are not observed outside of the select 13 states for which we have access to LEHD wage income records. Online Appendix I provides evidence that differential migration is likely not driving our results.⁴⁶

TABLE VI.

Effect on Earnings and Employment

	1–4 quarters after filing				5–8 quarters after filing
		OLS	IV			OLS	IV
	(1)	(2)	(3)		(4)	(5)	(6)
Earnings	4,300	−229***	−323*		4,254	−269***	−613**
	(3,809)	(9)	(175)		(3,885)	(13)	(248)
	[374,400]				[336,396]
By location
Cook County	4,821	−286***	−445*		4,821	−320***	−627*
	(5,810)	(12)	(249)		(5,956)	(17)	(337)
New York	3,779	−172***	−201		3,687	−218***	−599*
	(4,926)	(14)	(245)		(4,991)	(19)	(363)
By group
Female	4,136	−195***	−504***		4,094	−238***	−767***
	(3,545)	(10)	(185)		(3,610)	(14)	(295)
Black	4,319	−199***	−377		4,252	−247***	−931***
	(3,664)	(12)	(234)		(3,718)	(16)	(307)
Employment	0.565	−0.013***	−0.015		0.549	−0.019***	−0.018
	(0.317)	(0.001)	(0.021)		(0.322)	(0.001)	(0.027)
	[376,400]				[340,396]
By location
Cook County	0.623	−0.012***	0.003		0.613	−0.014***	−0.010
	(0.432)	(0.001)	(0.027)		(0.438)	(0.002)	(0.030)
New York	0.507	−0.014***	−0.032		0.485	−0.024***	−0.027
	(0.465)	(0.002)	(0.032)		(0.471)	(0.002)	(0.046)
By group
Female	0.585	−0.013***	−0.036		0.568	−0.019***	−0.003
	(0.315)	(0.001)	(0.025)		(0.320)	(0.002)	(0.034)
Black	0.583	−0.011***	−0.059*		0.566	−0.018***	−0.089**
	(0.316)	(0.001)	(0.031)		(0.321)	(0.002)	(0.040)

Notes. This table reports equally weighted averages of Cook County and New York nonevicted sample means , as well as equally weighted averages of location-specific OLS (OLS), and two-stage least squares (IV) estimates of the effect of eviction on labor outcomes. Outcomes are listed on the left of each row. Results are shown for 1–4 quarters (columns (1)–(3)) and 5–8 quarters (columns (4)–(6)) after the eviction case is filed. Each panel shows results for a given outcome. Below the combined estimates in each panel we report estimates separately for each location and for the female and Black subsamples. “Earnings” are average quarterly wage income from our labor market data described in Section III. “Employment” is the share of quarters with positive wage income from our labor market data described in Section III. Controls for all model specifications are the same as those described in Table IV. In each regression, we also control for tenants’ earnings and employment in each of the four quarters before filing, as well as averaged values over the eight quarters (two years) prior to the case filing. Standard errors for regression model coefficients are included in parentheses and are clustered at the judge(courtroom)-year level. Observation counts for the main combined specifications are reported in brackets below the standard errors in columns (3) and (6). Observation counts for all regressions shown here can be found in Online Appendix Table H.15. The reduced-form results for regressions shown above can be found in Online Appendix Table H.12. Cook County results were approved for release by the U.S. Census Bureau, authorization no. CBDRB-FY22-072. * p < .1, ** p < .05, *** p < .01.

Turning to employment, the IV estimate shows that for marginal tenants, eviction causes a 1.5 percentage point reduction in employment 1–4 quarters after filing, though this estimate is not statistically significant. The OLS estimate is statistically significant and similar in magnitude, suggesting that evicted tenants have employment rates that are 1.3 percentage points lower than nonevicted tenants. The IV point estimates in quarters 5–8 are similar and remain statistically insignificant. In contrast, the subgroup estimates suggest that eviction decreases Black employment by 8.9 percentage points (15% of the nonevicted mean), which is statistically significant but somewhat imprecisely estimated. Nevertheless, we can reject a test of equality of effects for Black and non-Black tenants at conventional levels (see Online Appendix Table H.1).

Online Appendix Table H.6 shows that the longer-run effects of eviction on earnings and employment (quarters 9–16 and 17–24 after filing) are for the most part smaller in magnitude, though estimated with somewhat less precision. We can rule out effects larger than an 837 reduction in quarterly earnings in quarters 9–16 after filing with 95% confidence.

Our results show that eviction causes reductions in earnings in the first two years after the case, consistent with the disruptive effects of eviction on housing stability described previously. Perhaps the closest prior research on earnings is based on the Milwaukee Area Renters Study and matched comparisons of renters who report experiencing a forced move in the past two years to those who do not. Desmond and Gershenson (2016) report that a forced move increases job loss by 11–22 percentage points, depending on the specification and estimation method. Our analysis differs on a number of dimensions. First, our treatment is an eviction order rather than the broader category of forced moves, which includes court-ordered evictions but also informal evictions, landlord foreclosures, and housing condemnations. Second, we study tenants in eviction court rather than tenants in low-income neighborhoods. In contrast to Desmond and Gershenson (2016), we find more modest effects on quarterly employment. Nevertheless, we find economically meaningful effects on quarterly earnings, which could capture shorter unemployment spells and which to our knowledge has not been studied. These effects on earnings and employment are concentrated in the first two years after filing, when housing disruptions are also the most pronounced.

VI.D. Financial Health

Next we examine the effects of eviction on financial health and present these results in Table VII. The first row reports estimates of the effect of eviction on our index of overall financial health, and the remaining panels report impacts on each outcome that is used to construct the financial health index.⁴⁷

TABLE VII.

Effect on Financial Health

	1–4 quarters after filing				5–8 quarters after filing
		OLS	IV			OLS	IV
	(1)	(2)	(3)		(4)	(5)	(6)
Financial health index	−0.054	−0.107***	−0.107*		0.008	−0.103***	−0.141
	(0.737)	(0.005)	(0.060)		(0.747)	(0.005)	(0.094)
	[269,814]				[271,230]
By location
Cook County	−0.075	−0.127***	−0.202**		−0.027	−0.130***	−0.230
	(0.990)	(0.008)	(0.102)		(1.002)	(0.009)	(0.174)
New York	−0.032	−0.087***	−0.012		0.043	−0.077***	−0.053
	(1.091)	(0.005)	(0.063)		(1.108)	(0.006)	(0.072)
Credit score	547.59	−8.40***	−7.86		551.84	−7.99***	−16.53**
	(67.11)	(0.38)	(5.18)		(68.61)	(0.41)	(6.67)
By location
Cook County	531.94	−9.19***	−8.69		536.62	−9.40***	−24.16**
	(74.04)	(0.55)	(8.29)		(74.56)	(0.59)	(11.15)
New York	563.24	−7.60***	−7.03		567.06	−6.58***	−8.90
	(111.94)	(0.53)	(6.21)		(115.19)	(0.56)	(7.33)
No open revolving account	0.481	0.032***	−0.039		0.468	0.037***	0.052
	(0.334)	(0.002)	(0.025)		(0.331)	(0.003)	(0.051)
By location
Cook County	0.587	0.032***	−0.072*		0.589	0.034***	0.080
	(0.491)	(0.003)	(0.043)		(0.491)	(0.005)	(0.099)
New York	0.375	0.032***	−0.006		0.347	0.041***	0.024
	(0.452)	(0.002)	(0.025)		(0.445)	(0.002)	(0.027)
Total balance: collections and delinquencies	2,550	153***	310		2,378	44*	548
	(4,099)	(24)	(393)		(3,936)	(25)	(502)
By location
Cook County	2,759	54	735		2,516	−34	739
	(5,504)	(36)	(659)		(5,291)	(38)	(930)
New York	2,342	253***	−115		2,240	122***	357
	(6,075)	(32)	(428)		(5,829)	(31)	(377)
Any auto loan or lease	0.170	−0.021***	−0.061**		0.176	−0.025***	0.031
	(0.264)	(0.001)	(0.030)		(0.269)	(0.002)	(0.036)
By location
Cook County	0.197	−0.040***	−0.130**		0.198	−0.047***	0.037
	(0.396)	(0.002)	(0.054)		(0.397)	(0.003)	(0.066)
New York	0.142	−0.002	0.008		0.155	−0.004**	0.025
	(0.349)	(0.002)	(0.027)		(0.362)	(0.002)	(0.026)

Notes. This table reports equally weighted averages of Cook County and New York nonevicted sample means , as well as equally weighted averages of lagged dependent variable OLS (OLS) and two-stage least squares (IV) estimates of the effect of eviction on outcomes related to the tenant’s financial health. Outcomes are listed on the left of each row. Results are shown for 1–4 quarters (columns (1)–(3)) and for 5–8 quarters (columns (4)–(6)) after the eviction case is filed. Each panel shows results for a given outcome. Below the combined estimates in each panel, we report estimates separately for each location. “Financial health index” is the average of any observed values of the index during the listed quarters. The index is an equally weighted index of the attributes listed below, and described in Section III. “Credit score” is the average of observed Vantage scores during the listed quarters. “No open revolving account” is the average of indicators for having no open revolving account over the listed quarters. “Total balance: collections and delinquencies” is the average of the balance in delinquent accounts or in collections over the listed quarters. “Any auto loans or lease” is an indicator for whether the individual is observed with an auto loan or lease in any of the listed quarters. Controls for all model specifications are those described in Table IV, except we do not control for race, which is not included in the data provided by the credit bureau. We also control for lagged values of the relevant outcome for up to two years prior to filing. Standard errors for regression model coefficients are included in parentheses and are clustered at the judge(courtroom)-year level. Observation counts for the main combined specifications are reported in brackets below the standard errors in columns (3) and (6), in the top panel. Observation counts for all regressions shown above can be found in Online Appendix Table H.15. The reduced-form results for regressions shown above can be found in Online Appendix Table H.13. * p < .1, ** p < .05, *** p < .01.

Eviction worsens tenants’ financial health, reducing the financial health index by 0.11 std. dev. in quarters 1–4 after filing according to IV, which is marginally significant, and by 0.10 std. dev. according to OLS. During this period, we find that eviction reduces the probability of having any auto loan or lease, which may be viewed as a proxy for durable-goods consumption (Dobkin et al. 2018; Agarwal et al. 2022), by 6.1 percentage points (36% relative to the nonevicted group mean), which is driven entirely by Cook County. The other point estimates during the first year imply reductions in credit access and increasing debt, but none of the estimates are individually significant. In quarters 5–8, the point estimate for effects on the financial health index are even more negative but also less precise and not significant. Eviction reduces credit scores by 16.5 points in this period.⁴⁸ By reducing credit scores, eviction could lead to increased borrowing costs for tenants and, to the extent that landlords use credit scores to screen tenants, hamper tenants’ ability to secure new housing.

The negative effects of eviction on financial health are more pronounced in the longer run. In Online Appendix Table H.7, we report estimates for effects in quarters 9–16 and 17–24 after filing. Eviction reduces the composite index by 0.21 and 0.26 std. dev. in years 3–4 and 5–6, respectively, both of which are statistically significant at the 5% level. In quarters 9–16 after filing, eviction lowers credit scores (IV estimate of −16.8) and increases balances in delinquent accounts (IV estimate of 847).⁴⁹ In quarters 17–24, eviction increases the probability of having no open source of revolving credit (IV estimate of 9.3 percentage points, p < .10) and decreases the likelihood of having an auto loan or lease (IV estimate of 8.3 percentage points, p < .10). For both balances in delinquent accounts and credit scores, the IV estimate is larger than the OLS estimate, suggesting that compliers are more likely to be on the margin of having access to conventional credit sources.

Taken together, these results suggest that eviction causes further deterioration in tenants’ financial circumstances and reduces subsequent access to credit. We find reductions in the financial-health index that are comparable to the effect of having a Chapter 13 bankruptcy filing dismissed (Dobbie, Goldsmith-Pinkham, and Yang 2017). Our estimated effects on credit scores are similar in magnitude to the effect that moving to a low-poverty neighborhood has on children’s future credit scores (Miller and Soo 2021), or the effect of removing a bankruptcy flag from a credit report (Gross, Notowidigdo, and Wang 2020; Dobbie et al. 2020). In contrast to the effects on housing, homelessness, and labor market outcomes documented above, the effects on financial health are larger in the longer run.

VI.E. Hospital Visits

We investigate the effects of eviction on hospital use in New York, where we have access to hospital data. Table VIII reports estimates for three measures of hospital use: the total number of nonpregnancy-related hospital visits, the total number of emergency room visits, and the total number of hospital visits for mental health conditions. The table includes results for the first and second year following a case.

TABLE VIII.

Effect on Hospital Use

	1–4 quarters after filing				5–8 quarters after filing
		OLS	IV			OLS	IV
	(1)	(2)	(3)		(4)	(5)	(6)
Number of hospital visits	0.739	0.055***	0.188**		0.632	0.039***	−0.113
	(1.321)	(0.006)	(0.094)		(1.208)	(0.006)	(0.142)
Number of emergency visits	0.588	0.045***	0.106		0.511	0.028***	−0.065
	(1.091)	(0.005)	(0.089)		(1.010)	(0.005)	(0.124)
Number of mental health visits	0.047	0.016***	0.054*		0.045	0.012***	−0.035
	(0.295)	(0.001)	(0.030)		(0.346)	(0.002)	(0.055)
	[179,024]				[154,531]

Notes. This table reports the effects of eviction on hospital use in New York. The table includes the nonevicted sample means , OLS (OLS) estimates and two-stage least squares (IV) estimates of the effect of eviction on hospital use. Outcomes are listed on the left of each row. Results are shown for 1–4 quarters (columns (1)–(3)) and 5–8 quarters (columns (4)–(6)) after the eviction case is filed. “Number of hospital visits” is the total number of nonpregnancy-related hospital visits. “Number of emergency visits” is the total number of nonpregnancy-related emergency room visits, and “Number of mental health visits” is the total number of nonpregnancy-related hospital visits for mental health conditions, where mental health conditions are defined in Online Appendix Table C.4. Controls for all model specifications are the same as those described in Table IV and lagged values of the number of hospital visits, and visits by diagnosis type. Standard errors are included in parentheses and are clustered at the courtroom-year level. Observation counts for all outcomes are listed at the bottom of the table, in brackets. The reduced-form results for regressions shown above can be found in Online Appendix Table H.14. * p < .1, ** p < .05, *** p < .01.

In the first year after the case, eviction increases total hospital visits by 0.19 visits in the first year following the case (29% relative to the nonevicted mean). Estimates for the total number of emergency room visits are similar in magnitude, although they are not statistically significant. Eviction also increases the number of visits to a hospital for mental health conditions by about 0.05 visits in the first year, a more than 100% increase over the nonevicted mean.⁵⁰ In the second year after the case, the IV estimates are insignificant and less precise. We explore longer-run effects on hospital use in Online Appendix Table H.8, where results remain statistically insignificant and imprecise. Compared to the IV estimates, OLS estimates tend to be somewhat smaller in the first year and somewhat larger in later years.

Overall, the effects of eviction on hospital use appear concentrated in the period shortly after the case filing. The finding that eviction causes increases in hospital visits is consistent with evidence from Currie and Tekin (2015), who find that foreclosures increase trips to the hospital. These impacts may reflect a deterioration in tenants’ health, but they may also reflect the use of hospitals as an alternative temporary source of shelter.⁵¹

VI.F. Comparisons across Locations

Figure VI plots estimates for Cook County on the vertical axis and estimates for New York on the horizontal axis. We standardize all estimates by multiplying the regression coefficient by the standard deviation of the eviction indicator and dividing by the standard deviation of the outcome. Across outcomes, the 2SLS estimates are similar across locations, with many of the estimates falling close to the 45-degree line, and with very similar estimates for employment and earnings across locations. The effects on some financial health outcomes are somewhat larger in Cook County, with statistically significant differences in having an auto loan or lease one year after the case. This may partially be driven by higher rates of car ownership in Cook County.⁵² The effects on residential mobility are somewhat larger for New York, which is consistent with New York’s lower vacancy rate and may be driven by fewer nonevicted tenants choosing to leave in the year or two after the case in New York. Effects on homelessness outcomes are also somewhat larger for New York, which is again consistent with a tighter housing market and may also stem from New York’s more extensive homeless shelter network and right to shelter law.

Figure VI.

Comparing Estimates across Locations

This figure plots standardized 2SLS estimates for Cook County (y-axis) against standardized estimates for New York (x-axis). All coefficients have been standardized by multiplying by the ratio of the standard deviation of the fraction evicted to the standard deviation of the outcome. Results are for one and two years after the case filing. Colors represent different outcomes, while the shapes indicate different outcome groups: housing, labor market, and financial outcomes. Cook County results were approved for release by the U.S. Census Bureau, authorization no. CBDRB-FY22-072. Circles report IV estimates, and triangles report OLS estimates.

VI.G. What about Noncomplier Cases?

The IV estimates can be interpreted as a weighted average of causal effects of eviction for compliers. One might additionally be interested in whether these estimates are similar to effects for the full population of evicted tenants. One possible approach to drawing inference about these effects is DiD. However, as discussed in Section IV, there are differential pretrends between evicted and nonevicted tenants for several outcomes in our settings, raising concern about the parallel-trends assumption.⁵³  Heckman and Robb (1985) and Ashenfelter and Card (1985) show that if shocks to outcomes follow a transitory and covariance stationary process—relatively strong assumptions in our setting—the DiD estimator will be unbiased when the pre- and postperiod are chosen symmetrically around the treatment period, even when the parallel-trends assumption does not hold.⁵⁴ In Online Appendix J, we further develop the symmetric DiD approach and compare symmetric DiD estimates to IV estimates which, under the appropriate assumptions, allow us to compare the ATT to the IV estimates.

Online Appendix Tables J.1–J.4 compare the IV estimates to symmetric DiD estimates. The symmetric DiD estimates for housing outcomes are quite similar to the IV estimates, while the effects for residential mobility are somewhat smaller. For labor market and financial health outcomes, DiD estimates have the same sign but tend to be smaller in magnitude. For health-related outcomes, the DiD and IV estimates both point to sizable increases in hospital use in the year after filing, but DiD estimates remain positive and statistically significant in the second year. Overall, the DiD estimates consistently show results that are broadly similar to the IV estimates but are smaller in magnitude, suggesting that the effects for the average evicted tenant are smaller than those for the marginal tenant.

VII. Conclusion

Evictions are a widespread phenomenon in the U.S. housing market, affecting more than two million households each year who overwhelmingly reside in poor or minority neighborhoods. Growing concern over evictions has spurred governments to pursue policies to reduce their incidence, citing substantial costs to tenants and local governments in the fallout from eviction. Such policies include legal aid to tenants facing eviction, emergency rental assistance, and just-cause eviction laws. Despite the large number of evictions and the growing policy interest, the consequences of eviction are not well understood. We explore how eviction affects tenants in housing court using newly linked administrative data from two large urban areas and a quasi-experimental research design that enables us to isolate causal effects of eviction.

We document signs of increasing economic distress in the lead-up to case filing across a broad range of measures: falling earnings, decreased attachment to the labor market, rising unpaid bills, and increases in hospital visits. This suggests many eviction cases are precipitated by adverse events. As we show, these patterns are likely to bias both comparisons of evicted tenants to renters outside of court and comparisons of evicted to not-evicted tenants in court, underscoring the value of our IV design that uses the random assignment of judges to estimate the effect of an eviction order for complier cases.

Using our IV design, we find that eviction exacerbates the economic distress experienced by tenants in the lead-up to a court filing. In the two years after a case, eviction increases homelessness, residential mobility, and hospital visits. During this period of disruption, eviction also reduces earnings, with particularly large effects for female and Black tenants. In the longer run, eviction worsens financial health through reduced credit scores and increased indebtedness.

This research speaks to an active policy debate on how, if at all, governments should address evictions. Although aspects of the ongoing debate over eviction-related policies, such as the extent of general-equilibrium effects, remain unsettled, we make significant progress on the key question of whether and how eviction affects tenants. Our results suggest that averting an eviction order may yield considerable benefits for tenants. Beyond the reductions in earnings and worsened credit, the increases in hospital visits and use of homelessness services suggest that eviction affects physical, mental, and material hardship. The high cost to local governments of providing health care and homelessness services (Evans, Philips, and Ruffini 2019) imply that there are also considerable spillover costs for society from eviction. These costs are important inputs to evaluating eviction-related policies.

Data Availability

The code underlying this article is available in the Harvard Dataverse, https://doi.org/10.7910/DVN/Y9TZQI (Collinson et al. 2023)