A comprehensive demographic profile of the US evicted population

Significance

Each year 2.7 million households receive an eviction filing, but because court records only contain names and addresses, we know very little about the individuals and families affected. By linking millions of eviction court records to Census Bureau data, we provide the most comprehensive description to date of the population of US renters facing eviction. Each year, 2.9 million children are affected by an eviction filing, and the typical eviction case filed in America involves one child. Further, we find enormous racial disparities in filing and eviction rates that persist across levels of income. Our findings have direct policy relevance as documenting these racial disparities in eviction is a precondition for establishing disparate impact claims under the Fair Housing Act.

Abstract

Millions of American renter households every year are threatened with eviction, an event associated with severe negative impacts on health and economic well-being. Yet we know little about the characteristics of individuals living in these households. Here, we link 38 million eviction court cases to US Census Bureau data to show that 7.6 million people, including 2.9 million children, faced the threat of eviction each year between 2007 and 2016. Overall, adult renters living with at least one child in their home were threatened with eviction at an annual rate of 10.4%, twice that of adults without children (5.0%). We demonstrate not only that the average evicted household includes one child, but that the most common age to experience eviction in America is during childhood. We also find that previous studies have underestimated racial disparities in eviction risk: Despite making up only 18.6% of all renters, Black Americans account for 51.1% of those affected by eviction filings and 43.4% of those evicted. Roughly one in five Black Americans living in a renter household is threatened with eviction annually, while one in ten is evicted. Black–White disparities persist across levels of income and vary by state. In providing the most comprehensive description to date of the population of US renters facing eviction, our study reveals a significant undercount of individuals impacted by eviction and motivates policies designed to stabilize housing for children and families.

One-third of households in the United States (~44 million households)^* rent their homes. Each year, 2.7 million of these households are threatened with removal through the legal process of court-ordered eviction (1). Eviction is the precipitating cause of a wide range of economic, physical, and mental hardships (2–5), all of which have been exacerbated by the COVID-19 pandemic (6).

Court records allow researchers and policymakers to estimate the number of households facing eviction, but they provide no description of the characteristics or composition of these households. Eviction filings contain no information on tenants beyond name and address. The most comprehensive estimates of individual-level demographic correlates of eviction rely on local surveys of specific cities (7) or statistical algorithms that impute the race/ethnicity and gender of defendants from the names and addresses listed in court records (8, 9). However, these imputation strategies are prone to errors, particularly when estimating defendant race/ethnicity, and may result in conservative estimates of disparities in eviction (10).

Even if imputation strategies were entirely reliable, they would only allow us to observe the race/ethnicity and gender of those directly listed on eviction filings—typically the leaseholders—overlooking all other household members not listed, including children. The median eviction case includes a single listed defendant, but the typical renter household has 2.4 members. Notably, many renter households include at least one child under the age of 18, and local studies have found that the presence of children in a household is associated with heightened eviction risk (7, 11).

Owing to these data limitations, previous studies have a) drastically underestimated the population threatened^† with court-ordered eviction each year, b) been unable to account for the number of children annually at risk of displacement, and c) been limited in rigorously documenting how eviction rates vary by race/ethnicity, gender, nativity, and age. These limitations have implications for understanding disparities and informing public policy.

The present study overcomes these limitations and makes three significant contributions to our understanding of housing precarity in the United States. First, we reveal the full population facing eviction each year, including millions of adults rendered invisible in eviction court records. Second, we reveal the incidence of eviction among children across the country, finding that 2.9 million children face the threat of eviction each year. The number of children experiencing housing instability vastly outstrips previous estimates based on homelessness data alone (12). Third, we provide the most accurate and comprehensive estimates to date of demographic variation in eviction risk, documenting significant racial disparities. Our findings have direct policy relevance, as documenting such disparities is a precondition for establishing disparate impact claims under the Fair Housing Act (13).

Linked Dataset of Eviction Records

We created a unique dataset linking eviction court records to the 2006 to 2015 American Community Survey (ACS), allowing us to observe detailed household rosters for households filed against for eviction. We draw on court records compiled by the Eviction Lab at Princeton University (14). These records, comprising 58 million filings from 2000 to 2016, were collected either manually or via bulk extracts from administrative data systems. They were cleaned, stripped of duplicate and commercial eviction cases, geocoded, and validated against publicly available data sources published by county and state court systems (15). We submitted these records to the US Census Bureau’s Person Validation Identification System (PVS), which then assigned Protected Identification Keys (PIKs) using a probabilistic linkage between records based on names and addresses reported in eviction filings (38 million matches; 65% PIK match rate nationally). We merged eviction records to the ACS by PIK, linking ACS responses in the year immediately preceding the eviction filing to observe household characteristics at approximately the time of the court case. This merged sample covered filings from 2007 to 2016 linked to ACS characteristics from 2006 to 2015 (N = 214,000 unique individuals whose household responded to the ACS in the year prior to being filed against). ACS responses allowed us to record the race, ethnicity, gender, nativity, and age of all members of households threatened with eviction. All statistics are weighted using the ACS sample weights. Additionally, we compared a merged sample of 2011 filings to 2010 Decennial Census data; see SI Appendix for more details.

To estimate eviction filing and judgment rates, we first calculated ratios per unique household filed against in our matched sample (e.g., the average number of US-born White men aged 40 to 45 per household filed against) by state and year. We then multiplied these ratios by publicly available estimates of total unique households filed against in every state-year (1) to calculate complete numerators: the total number of unique renters filed against and/or evicted by state, year, race, ethnicity, gender, nativity, and age (SI Appendix). This dataset allows us to describe the total population at risk of eviction (those living in a household that received an eviction filing), which is composed of listed adults (those directly filed against), unlisted adults (those living in the household but unnamed on the filing), and children (those under 18 and typically unnamed on the filing). We similarly describe the total evicted population (those living in a household that received an eviction judgment: a court order to vacate the premises).

Results

Between 2007 and 2016, an average of 7.6 (95% CI: 6.9 to 8.3) million individuals faced the threat of eviction, and 3.9 (3.5 to 4.2) million were evicted each year (Fig. 1A). This population includes those listed in court records as well as other adults and children living in the household. Of those who faced the threat of eviction, only 3.2 (2.9 to 3.5) million were listed in court records. Relying on such records alone, which researchers and policymakers regularly do, thus underestimates the annual population at risk of eviction by 57.9%.

Fig. 1.

Revealing the total population affected by eviction. (A) The total average annual (2007 to 2016) population filed against and evicted is composed of adults listed on filings, unlisted adults living in the household, and children under 18 living in the household. (B) The distribution of relationships between unlisted adults and the reference adult listed on the filing. All estimates include 95% CI. Data are based on 2007 to 2016 eviction records linked to the 2006 to 2015 ACS. Census Disclosure Review Board Approval Number: CBDRB-FY23-CES004-012.

Of the 7.6 million individuals facing eviction each year, nearly 40%—2.9 (2.7 to 3.2) million—were children. The average eviction case filed in America involved roughly one child under age 18 (SI Appendix, Fig. S1A). Children were present in 52.2% of renter households filed against compared to only 33.5% of renter households not filed against (SI Appendix, Fig. S1B). Of the 7.6 million individuals facing eviction each year, 1.5 (1.3 to 1.6) million were unlisted adults. We examined the relationship of these unlisted adults to the ACS reference person, who was listed in most filings (Fig. 1B). Unlisted adults were most commonly the adult children of those filed against (41.7%) followed by spouses (19.0%) and unmarried partners (13.4%).

Non-Hispanic Black renters were the only race/ethnicity group overrepresented in eviction filings and judgments. Fig. 2 summarizes the distribution of the population threatened with eviction and compares this to the distribution of the total renting population by race/ethnicity and gender. Overall, Black Americans made up only 18.6% of all renters yet accounted for 51.1% of those threatened with eviction and 43.4% of those who were evicted. By contrast, although White Americans make up just over half of all renters (50.5%), they accounted for only 26.3% of those threatened with eviction and 32.0% of those who were evicted.

Fig. 2.

Black renters are disproportionately evicted. Average annual (2007 to 2016) proportions by race/ethnicity and gender of the total population living in renter households, the total population living in households filed against, and the total population living in households evicted. All estimates include 95% CI. Data are based on 2007 to 2016 eviction records linked to the 2006 to 2015 ACS. Census Disclosure Review Board Approval Number: CBDRB-FY23-CES004-012.

Those living with children faced much higher risks of being threatened with eviction and being evicted than those without children (Fig. 3). Adult renters living with at least one child in their home were threatened with eviction at an annual rate of 10.4%, compared to 5.0% for those without children. This gap was particularly pronounced for Black women, where filing rates were 28.4% with children present and 16.3% for those without children. Overall, our estimates indicate that, between 2007 and 2016, roughly one in five Black adult renters were living in a household filed against for eviction and roughly one in ten were evicted each year. By contrast, the average annual eviction filing and eviction rates for White adult renters were 4.2% and 2.5%, respectively. Eviction filing and eviction rates for Hispanic adult renters were comparable to those for White renters. Asian renters consistently had the lowest eviction filing and eviction rates^‡.

Fig. 3.

Racial disparities in filing and eviction rates across all adult renters. (A) Average annual (2007 to 2016) filing and eviction rates for all adult renters by race/ethnicity and gender in households with children under 18 and (B) households with no children. Numerators include both listed and unlisted adults. All estimates include 95% CI. Data are based on 2007 to 2016 eviction records linked to the 2006 to 2015 ACS. Census Disclosure Review Board Approval Number: CBDRB-FY23-CES004-012.

Previous research has suggested a higher risk of eviction for women (8). While we find some significant gender disparities within race/ethnicity and the presence of children for the total population affected by eviction (Fig. 3), these differences were relatively small. For example, Black women and Black men living with children had eviction rates of 12.1% (11.7 to 12.4%) and 10.9% (10.5 to 11.4%), respectively. However, there are substantial gender disparities when only focusing on the listed population (SI Appendix, Fig. S2). We show that a large driver of the total gender disparity is the disproportionate listing of Black women on eviction filings compared to Black men, especially in households with children. In addition to race/ethnicity, gender, and the presence of children, we found that nativity is an important dimension of eviction risk. Within race/ethnicity, filing and eviction rates were roughly twice as high among US-born renters relative to foreign-born renters (SI Appendix, Fig. S3).

Fig. 4A presents the age distribution of the full renter population and the populations threatened with eviction and evicted. Just over two-thirds of American renters (67.2%) were below the age of 40, but more than three-quarters (77.2%) of those facing the threat of eviction were below that age. Fully 32.9% of the population threatened with eviction was below age 15. The most common age to be evicted in America was between birth and 19 y old (Fig. 4B). Of all children aged 0 to 4 living in renting households, 5.7% (5.5 to 5.8%) were evicted each year; for Black children aged 0 to 4, 12.4% (11.9 to 13.0%) were evicted annually (SI Appendix, Fig. S4). The filing rate for renters above age 65 was 2.0%, amounting to roughly 168,000 elderly renters facing eviction each year.

Fig. 4.

The age distribution of eviction. (A) Average annual (2007 to 2016) proportions of the total population filed against and evicted by 5-y age groups and (B) average annual (2007 to 2016) filing and eviction rates by 5-y age groups. Dotted bars (A) show proportions of the total renter population in each 5-y age group for comparison. All estimates include 95% CI. Data are based on 2007 to 2016 eviction records linked to the 2006 to 2015 ACS. Census Disclosure Review Board Approval Number: CBDRB-FY23-CES004-012.

We further examine eviction patterns across time, states, and household income. We show that eviction rates have been remarkably stable over time in all groups (SI Appendix, Fig. S5), vary significantly by state (SI Appendix, Figs. S6 and S7), and decrease as income increases (Fig. 5). Still, eviction rates remain much higher for Black renters even at higher income levels.

Fig. 5.

Racial disparities in eviction rates by income. (A) Average annual (2007-2016) eviction rates for all adult renters by race/ethnicity, gender, and absolute inflation-adjusted 2018 income in households with children under 18 and (B) households with no children. Similar rates are reported with household income categorized as a percentage of county median household income for all adult renters in (C) households with children under 18 and (D) households with no children. Numerators include both listed and unlisted adults. All estimates include 95% confidence intervals. Data are based on 2007-2016 eviction records linked to the 2006-2015 ACS. Census Disclosure Review Board Approval Number: CBDRB-FY23-0407.

Discussion

Reliable and transparent data systems are necessary for exposing ongoing inequities. For example, despite the enactment of the Fair Housing Act (FHA) more than 50 y ago, studies continue to find persistent ethno-racial disparities in the home mortgage market (16, 17). These studies are only possible because of the 1975 Home Mortgage Disclosure Act (HMDA) requiring lenders to report loan-level data that includes borrower characteristics relevant to the FHA (e.g., race/ethnicity). While it has remained difficult to bring disparate impact cases forward under the FHA, the problem is at least made visible due to transparent, mandatory reporting standards.

There is no policy like the HMDA for the private rental market, where the vast majority of low-income Americans—especially Black and Hispanic Americans—find housing. As we demonstrate here, the lack of reporting standards in housing courts has inhibited our understanding of the full scope of eviction, including the prevalence of eviction in the lives of children and massive racial disparities in eviction rates owing to the ongoing history of discriminatory housing policies and practices.

Over twice as many adults and children are threatened with eviction in an average year than previously found in studies exclusively examining the names of lessees in court ledgers (1). Black renters experience far and away the highest risk of eviction. Notably, our estimates are likely still conservative, as Black individuals are less likely to receive a PIK compared to White individuals (18). We show that the gender gap in eviction, previously documented in studies drawing on court records, is largely driven by the higher likelihood that Black women are listed in eviction filings compared to Black men. This may be due to Black women being more likely to be included on apartment leases. Still, the gender disparity in filings is significant given that being listed on an eviction case can have negative consequences above and beyond living in a household threatened with removal, including through credit and screening systems that make securing future housing difficult. In this sense, Black women face a double burden: They are most likely to live in a household targeted for eviction (especially if children are present) and, within that household, are most likely to bear the mark of the eviction court record itself.

Our finding that foreign-born renters faced court-ordered evictions much less often than their US-born peers must be interpreted with caution, as foreign-born renters are also less likely to receive a PIK (18). However, the magnitude of the difference is such that it is unlikely to be explained by matching variation alone. It is possible that this disparity is not driven by lower rates of displacement, but rather lower rates of court-ordered eviction. It may be that some foreign-born renters have limited access to the legal system and are then disproportionately subjected to informal or illegal evictions that are not reflected in court records.

The presence of children in a household significantly increases the risk of eviction. Across the life course, the risk of experiencing an eviction is highest during childhood. Further, we show that almost half of unlisted adults threatened with eviction are the adult children of those filed against. This highlights the intergenerational nature of eviction, including the difficulty of transitioning to independent housing after growing up in a low-income renting household. Being evicted is a traumatic event, especially in early life. Evicted children face increased risk of food insecurity (19), exposure to environmental hazards (20), academic challenges (21), and a range of long-term physical and mental health problems (5). Previous research on housing instability among children has primarily focused on the prevalence of homelessness, with estimates of 1.27 million homeless children enrolled in public schools in the academic year 2019 to 2020 (12). We demonstrate that over twice that number are at risk of losing their homes in any given year, motivating policies aimed at helping to stabilize children and families.

Future research will be needed to explore how disparities in eviction risk by race/ethnicity might be mediated by economic conditions and policy context. By constructing a linked dataset, we establish the scope of the American eviction crisis—including in the lives of children, 2.9 million of whom face eviction each year; and of Black Americans, who face substantially higher eviction rates across the income distribution—which can motivate research agendas and inform policy priorities.

Materials and Methods

We draw on eviction court records compiled by the Eviction Lab at Princeton University (14). Studies based on court-ordered eviction records produce more accurate estimates than those reliant on self-reports in surveys (11). Administrative data from court systems contain millions of records but limited information about each case: names of plaintiffs (e.g., landlords, property managers) and defendants (tenants), defendant addresses, and action dates. These data are sufficient to enable linkage to other records that have demographic information, like the ACS, and thereby to examine eviction outcomes across subpopulations.

Our analyses are based on the sample of filings that we can link to PIKs and that responded to the ACS in the year immediately prior to being filed against. This introduces two sources of potential bias: The probability of being assigned a PIK is nonrandom and/or the probability of responding to the ACS prior to being filed against is nonrandom. Further, to the extent these probabilities vary systematically across our key dimensions, our comparisons may reflect selection differences rather than true differences in eviction risk.

First, we tested estimating filing and eviction rates with a restricted sample of only county-years with higher PIK rates (>75%); at the state-level, PIK rates varied from 48% in Hawaii to 78% in Washington D.C. Results are substantively similar, except for the Native Hawaiian and Pacific Islander (NHPI) and the American Indian and Alaskan Native (AIAN) populations (SI Appendix, Fig. S8). To put these PIK rates in perspective, our study represents a relatively unusual case for the PVS process because we only have names and addresses in eviction filings. PVS typically assigns PIKs to data that also include date of birth, if not Social Security Numbers (22). The most similar PVS process to ours is a study in Cook County, Illinois, that was able to assign PIKs to 52% of individuals listed on eviction filings (3). A study linking bankruptcy filings to credit bureau records by name and address using a similar probabilistic matching algorithm had a match rate of 69% (23).

Second, we examined patterns using all filings in 2011 linked to the short-form 2010 Decennial Census where possible selection forces are likely much weaker; all results are virtually identical (SI Appendix, Fig. S9). We use the ACS sample in our main results in order to focus on representativeness over a wider time period (2007 to 2016). Third, we examined differences in ACS response rates. While the ACS is a stratified random sample of all American households, ACS response may be nonrandom among households that go on to be threatened with eviction the following year. To the extent that ACS response rates vary across these households by our key dimensions (e.g., race/ethnicity), our comparisons may be biased. We calculated county-year response rates to the ACS in the year prior to receiving an eviction filing using our entire sample of filings linked to PIKs. We then test whether these ACS response rates vary by county-level characteristics corresponding to our key dimensions using a linear regression model (SI Appendix, Table S1). The only statistically significant predictor of ACS response was the number of children in the household. However, the point estimate was extremely small and unlikely to affect our final results.

Because our sample of the total population affected by eviction only includes individuals living in households that contained at least one individual able to be matched to an eviction filing via PIKs, we are not able to directly estimate filing and eviction rates. Our eviction court records do not cover all eviction cases, and our PIK assignment is incomplete (65%). Rates calculated using our matched sample alone would be biased downward because while the denominators would be complete (i.e., all renters), the numerators would be too small (i.e., unique renters filed against in the matched sample).

We therefore indirectly estimated eviction filing rates by state (s), race (r), ethnicity (e), and sex (x), pooling over the period 2007 to 2016. To estimate rates, we first used our matched sample to estimate ratios of total unique renters filed against ( $f_{s,r,e,x}$ ) per unique household filed against ( $h_s$ ). We include both listed and unlisted renters filed against. We then multiplied these ratios by estimates of total unique households filed against taken from publicly available estimates covering every state and year from 2007 to 2016, which are adjusted for underreporting and incomplete eviction data ( $H_s$ ) (1). This yielded our target numerator: total unique renters filed against by state, race, ethnicity, and sex ( $F_{s,r,e,x}$ ). We then divided by our target denominator, total renters derived from the ACS ( $R_{s,r,e,x}$ ) to calculate filing rates by state, race, ethnicity, and sex. We repeated this exercise for eviction rates.

$$\frac{F_{s,r,e,x}}{R_{s,r,e,x}}=\frac{\frac{f_{s,r,e,x}}{h_s}\times H_s}{R_{s,r,e,x}}.$$

We estimated SE for these rates by bootstrapping, including uncertainty in the ratio estimates due to ACS sampling weights and uncertainty in the modeled estimates of unique households filed against ( $H_s$ ) reported as Bayesian credible intervals; see ref. 1 for details on estimation.

In scaling our ratio estimates in the matched sample by the modeled estimates of unique household filed against at the state-level ( $H_s$ ), we assume that our PIK-matched sample is representative of all households filed against within each state; this allows us to construct national estimates. We compare pooling information across counties within states in two ways. First, we pool households filed against across the state in calculating state-level ratios (Model 1; our preferred estimates). Second, we calculate county-level ratios scaled to county-level estimates of households filed against in counties where we have a relatively large sample of PIK-matched households filed against (>100; we refer to these as “covered” counties); in uncovered counties, we use state-level average ratios (Model 2). Alternatively, we can construct in-sample estimates by only using covered counties (Model 3). Comparing rates based on these different samples (SI Appendix, Fig. S10), we examine whether estimates are sensitive to extrapolation from covered counties to all counties in the state (Model 3 vs. Models 1 to 2) and how county-level information is pooled in this extrapolation (Model 1 vs. Model 2). SI Appendix, Fig. S10 illustrates very close alignment between these estimates, suggesting that national estimates are not sensitive to extrapolation within states based on our covered counties. We further examine the proportion of all households filed against by state ( $H_s$ ) that are located in covered counties (SI Appendix, Fig. S11) and compare county-level characteristics (e.g., household income) of covered vs. uncovered counties (SI Appendix, Fig. S12). These figures demonstrate that our PIK-matched sample covers almost all households filed against in each state and covered counties look very similar to uncovered counties on key characteristics. An exception is that we have relatively lower coverage of households filed against in New York and Pennsylvania (SI Appendix, Fig. S11), and uncovered counties may have a slightly higher proportion of White renters (SI Appendix, Fig. S12), though SI Appendix, Fig. S10 demonstrates that national estimates are not sensitive to extrapolation over these counties.

In supplementary results, we further examine whether eviction patterns vary by period, state, and income. First, we split our sample into three time periods: 2007 to 2009, 2010 to 2012, and 2013 to 2016. We find that filing and eviction rates for adult renters by race/ethnicity, gender, and the presence of children are all very stable over this period (SI Appendix, Fig. S5). Second, we estimate state-level filing and eviction rates for adult renters by race/ethnicity, gender, and the presence of children (SI Appendix, Figs. S6 and S7). We find substantial heterogeneity across states, especially in the Black–White disparity. We compare state-level rates estimated with the 2010 Census compared to the 2006-2015 ACS, and results are very similar (SI Appendix, Figs. S13 and S14). We also estimate the total number of children living in households filed against and evicted annually by state (SI Appendix, Fig. S15). Georgia, Texas, and New York threaten the highest number of children with eviction annually, while California evicts the most children. Compared to the national average of roughly one child per household evicted, Mississippi has the highest number of children per evicted household (1.41). Third, we report eviction rates for adult renters by race/ethnicity, the presence of children, and absolute/relative household income (Fig. 5). We demonstrate that eviction rates decrease as income increases. Still, eviction rates remain much higher for Black renters even at higher income levels. It is important to note that these linked data only allow us to observe income within 1 y of eviction filing (e.g., income observed in the 2010 ACS for a renter filed against in 2011). It is possible that eviction filings for relatively high household incomes are the result of an acute event (e.g., the loss of a job/spouse, medical emergency) that also led to a sharp decline in income between the point we are able to observe income and the eviction filing. Future research will be needed to explore how disparities in eviction risk by race/ethnicity might be mediated by economic conditions and policy context.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

All our empirical results use confidential microdata from the U.S. Census Bureau. We are not able to make these data directly available. However, we can provide the code and researchers can follow the directions on how to write a proposal to gain access to the data via a Federal Statistical Research Data Center using the Standard Application Process. The code to conduct our analyses is available here: https://github.com/ngraetz/pnas_demographics_of_eviction (24). To replicate our data and analyses, researchers would submit a common application through the Standard Application Process portal to apply for access to our confidential linked data from any of the 16 principal federal statistical agencies and units: https://www.census.gov/about/adrm/fsrdc.html.