Racial and Gender Disparities among Evicted Americans

Drawing on millions of court records of eviction cases filed between 2012 and 2016 in 39 states, this study documents the racial and gender demographics of America’s evicted population. Black renters received a disproportionate share of eviction filings and experienced the highest rates of eviction filing and eviction judgment. Black and Latinx female renters faced higher eviction rates than their male counterparts. Black and Latinx renters were also more likely to be serially filed against for eviction at the same address. These findings represent the most comprehensive investigation to date of racial and gender disparities among evicted renters in the United States.

F ORCED dislocation from housing is implicated in the reproduction of poverty and disadvantage. Residential eviction has been linked to a wide array of negative consequences, from homelessness and increased material hardship to depression and suicide (Desmond and Kimbro 2015;Osypuk et al. 2012). Documenting populations disproportionately at risk of eviction informs researchers, advocates, and policymakers striving to better understand and address disparities in access to stable housing. Such evidence may be critical in establishing the statistical basis for a prima facie case of a disparate impact claim under the Fair Housing Act (Schwemm and Bradford 2016).
Local studies have documented the demographic characteristics of evicted renters. The Milwaukee Area Renters Study found that eviction risk was higher for black and Latinx and lower-income renters, as well as those with children (Desmond and Gershenson 2017;Desmond, Gershenson, and Kiviat 2015;Desmond and Shollenberger 2015). Systematic review of names listed in eviction court records from Milwaukee County suggested that female renters-particularly in predominantly black and Latinx neighborhoods-were disproportionately evicted (Desmond 2012).
However informative, studies confined to a single city lack generalizability. Eviction is widespread-an estimated 1.6 million households nationwide are displaced annually (Desmond et al. 2018a)-yet no study has documented the demographics of America's evicted renters in national perspective. Are black and Latinx renters evicted at higher rates than their white counterparts? Are female renters evicted at higher rates than men, and is this true for all racial/ethnic groups? To address these questions, we drew on court records of eviction cases filed between 2012 and 2016 against roughly 4.1 million individuals in 39 states. We used these data to calculate a set of counts and rates that offer the most comprehensive examination to date of racial and gender disparities among evicted renters in the United States.

Data and Methods
We estimated eviction rates for men and women as well as multiple racial and ethnic groups. We drew on eviction records from 2012 to 2016, compiled by the Eviction Lab at Princeton University (Desmond et al. 2018a). These records were collected, either manually or via bulk extracts from court administrative data systems, by LexisNexis Risk Solutions. 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 (Desmond et al. 2018b). 1 We included in our sample any county for which the Eviction Lab could provide validated eviction records for at least one year between 2012 and 2016. In total, we observed 3,663 county-years from 1,195 unique counties, containing 37.5 percent of American renter households. Based on American Community Survey (ACS) five-year estimates for 2012 to 2016, these counties were close to representative of all counties along a number of key variables. 2 Court records provide a unique opportunity to examine the prevalence of eviction across time and space. Studies based on court-ordered eviction records produce more accurate estimates than those reliant on self-reports in surveys (Desmond 2012;Schwartz 1994). However, although administrative data from court systems contain millions of records, they provide limited information about each case. Records included case numbers, names of plaintiffs (e.g., landlords, property managers) and defendants (tenants), defendant addresses, and filing dates. Defendant gender and race/ethnicity were not recorded in eviction records.
Accordingly, we imputed demographic characteristics on the basis of defendants' names and addresses. We observed more than 4.1 million defendants listed in court records. We produced three predictions of defendant gender using the R packages gender (Mullen 2018) and genderizeR (Wais 2016), as well as the web service Gender API (Gender-API.com n.d.). 3 Drawing on defendants' first names, each method produced a prediction (0 to 1) that the defendant was female and the inverse probability that they were male. 4 We took the mean across all available predictions. Roughly 94.3 percent of names yielded more than one prediction, but the average variance between multiple predictions was extremely small (0.007).
To impute defendants' race/ethnicity, we used a Bayesian predictor algorithmthe wru package in R (Khanna, Imai, and Jin 2017)-that calculated race/ethnicity probabilities on the basis of two Census Bureau data sets: the Surname List and the 2010 Decennial Census. These data sets provide, respectively, the frequencies with which common surnames are associated with racial/ethnic groups and the racial/ethnic composition of each tract in the United States. Jointly, they allowed us to estimate the conditional probability of a defendant's race/ethnicity, given their surname and geolocation. Early attempts at racial imputation were prone to high error rates (Fiscella and Fremont 2006). However, Imai and Khanna's (2016) validation of predictions using the wru algorithm found that the combination of name and geolocation data resulted in much higher rates of correct classifications, compared with relying on names alone.
These imputation procedures allowed us to assign to each defendant a probability of being female or male and of being white, black, Latinx, Asian, or of another race/ethnicity. For each individual, the probabilities of belonging to each of the racial/ethnic groups summed to one, as did the probability of being female and male. 5 We multiplied gender probabilities by race/ethnicity probabilities, allowing us to categorize defendants by race/ethnicity and gender. Individuals were not assigned to a single race/ethnicity-by-gender category but given probabilities of falling into each. Assuming that cross-classified probabilities followed a multinomial distribution, we calculated the variance of each estimate. This approach allowed us to maintain and assess uncertainty inherent to the imputation process and to avoid misclassifications at the individual level.
We aggregated these probabilities within county-years to produce annual estimates of the number of individuals filed against and evicted in each cross-classified group (e.g., black women, white men). We summed variances, which allowed us to provide confidence intervals for these estimates. We also produced estimates that adjusted for serial eviction filings, cases in which property owners repeatedly file eviction cases against tenants at the same address, often to facilitate rent collection (Immergluck et al. 2020;Leung, Hepburn, and Desmond 2020). We linked cases that shared the same defendants and addresses within the same county-year and removed repeated filings to produce estimates of unique individuals filed against and evicted in each group. We then averaged these estimates across available county-years. 6 The resulting averages constituted the numerators in the rates we describe below. These counts reflected only those individuals who were listed as defendants in these cases, typically leaseholders (Desmond 2012). They omitted any additional adults who may have been living in the household but who were not formally contracted with the unit.
The denominators for many of these rates were counts of adult residents living in rental housing, also cross-classified by gender and race/ethnicity. The Census Bureau does not make such cross-classified counts directly available in a standard table form. Instead, we estimated the number of adult residents living in rental housing in each county using ACS five-year data for 2012 to 2016 from the Integrated Public Use Microdata Series (IPUMS) (Ruggles et al. 2019). IPUMS data allowed us to determine the race/ethnicity and gender of individuals older than 18 years who lived in rental housing and to weight these observations. 7 The drawback of the IPUMS data is that the smallest identifiable geographic unit is the Public Use Microdata Area (PUMA), whereas we sought to report eviction rates at the county level. In some cases, a PUMA corresponds to a single county; in other cases, a PUMA consists of several whole counties, contains a mixture of whole and partial counties, or is made up of several partial counties. To deal with the latter cases, we divided and aggregated the data according to the PUMAcounty geographic relationship. First, we downloaded tract-level counts of renting households by race/ethnicity of the household head, taken from ACS five-year estimates for 2012 to 2016. Second, using a tract-to-PUMA crosswalk, we aggregated these tract-level ACS data to the county-PUMA level, identifying Census tracts residing in the same county and PUMA. This allowed us to observe the fraction of renter household heads in each PUMA that belonged to a given county. We aggregated ACS-provided margins of error for tract-level estimates to calculate uncertainty around each of these fractions. Third, we split IPUMS PUMA-level data into constituent county-PUMAs based on the share and racial composition of renters observed in the previous step. Because ACS data did not allow us to observe gender ratios of renters within racial/ethnic categories at the county-PUMA level, we assumed that these ratios were the same between county-PUMAs in the same PUMA. We calculated uncertainty around cross-classified counts of renters under the assumption that the distribution in any given PUMA followed a multinomial distribution with n as the total number of renters. Once we obtained cross-classified counts of renting households at the county-PUMA level, we aggregated counts up to the county level to produce the necessary denominators.
Using these data, we calculated three statistics for every gender-by-race/ethnicity category. First, we report the eviction filing rate: the number of eviction filings divided by the renter population. An eviction filing is typically the first step in the eviction process recorded by the civil court system. Many tenants vacate their homes upon receipt of an eviction filing (Hartman and Robinson 2003). Even when they do so, having been filed against for eviction is marked in tenants' credit and rental history, limiting their future housing options and potentially damaging their credit. The eviction filing rates reported here are adjusted for serial eviction filings, counting only one instance of each serially filed case within each county-year. This adjustment allowed us to avoid double-counting individuals in the numerator. 8 Second, we report the eviction rate: the number of eviction judgments divided by the renter population. An eviction judgment is rendered by the courts when a case is decided in favor of the plaintiff (property owner or manager). The eviction rate is our best measure of the percentage of renters forcibly removed from their homes by court order. Eviction rate estimates are also adjusted for serial eviction filings, treating the outcome of the most recently observed case as final.
Third, we report the serial eviction filing rate: the number of individuals who are serially filed against divided by the total number of unique filing recipients. This rate allows us to assess whether certain demographic groups are at increased risk of being filed against repeatedly at the same address, a process that entails considerable financial costs because of late charges and legal fees that are shifted to tenants (Leung et al. 2020).
Rather than report statistics at the county level, which would give equal weight to small and large counties, we primarily provide estimates at the renter level. To do so, we sampled estimates from the county-level distribution specific to the race and gender of the renter. For example, Harris County, TX, was predicted to have 40,356 Asian male renters (standard error of 195.9) and 421 Asian male evictees, with a variance of 68.8. In the renter-level file we sampled 40,356 times from each of these distributions, calculating an eviction rate each time and thereby maintaining at the renter level the uncertainty inherent our county-level estimates. All figures and estimates reflect this uncertainty, and we include 95 percent confidence intervals when reporting count estimates. All estimates presented here are unconditional. Statistical analyses are limited to one-or two-tailed t-tests, which assess the differences in mean rates between groups. We have made our data and code publicly available at www.evictionlab.org/demographics-of-eviction-data. We hope researchers will use these data to conduct further analyses of the covariations between these rates and the sociodemographic, economic, and legal characteristics of counties.
The descriptive accounting of these rates and disparities between them-which we provide below-offers a precursor for such analyses.

Results
Across the 1,195 counties in our data, 1.44 million eviction cases were filed in an average year (including serial eviction filings), resulting in approximately 660,000 eviction judgments. As Figure 1 demonstrates, the shares of eviction filings and eviction judgments accruing to members of each racial/ethnic group were not proportional to their representation in the renter population across these counties. Black individuals were overrepresented in the evicted defendant population. They made up 19.9 percent of all adult renters but 32.7 percent of all eviction filing defendants. Four out of every five black renters in our sample (81.0 percent) lived in a county in which the share of eviction filings against black renters was higher than the share of the renting population that was black. All other racial/ethnic groups were underrepresented, with the largest absolute difference among white renters. White renters made up 51.5 percent of all adult renters but only 42.7 percent of all eviction filing defendants.
The overrepresentation of black renters within the population of renters against whom an eviction was filed is particularly apparent in highly populated counties. In Table 1 we list the 10 largest counties in our sample by total renter population. 9 For each, we record the share of the renter population and the share of filing recipients who were black, Latinx, and white. 10 In each of these counties, the share of filings against black renters was greater-often far greater-than their share of the renter population. In the most extreme case (King County, WA), blacks received 28.2 percent of all eviction filings, more than three times their share of the renter population (9.0 percent). White and Latinx renters were overrepresented among filing recipients in less than half of these counties, and the disparities were much smaller. On average across these 10 counties, the share of filing recipients who were black was 12.4 percentage points higher than the share of renters who were black. By contrast, the maximum overrepresentation in these counties for Latinx and white renters was 6.8 percentage points (Middlesex County, MA) and 4.8 percentage points (Tarrant County, TX), respectively.
After adjusting for serial eviction filings, the average renter faced a 4.1 percent eviction filing rate (median 3.6 percent) and an eviction rate of 2.3 percent (median 2.2 percent). Put another way, approximately one in 25 renters was threatened with eviction every year, and one in 40 was evicted.
Eviction filing and eviction rates varied considerably by race/ethnicity. Black renters experienced the highest average rates of eviction filing (6.2 percent) and eviction judgment (3.4 percent). By contrast, the average eviction filing rate among white renters was 3.4 percent, and the average eviction rate was 2.0 percent. Nearly one in four black renters (23.7 percent) lived in a county in which the black eviction rate was more than double the white eviction rate. Asian renters experienced the lowest rates, with an average eviction filing rate of 2.4 percent and an average eviction rate of 1.2 percent. The average Latinx eviction filing rate was 3.6 percent, significantly higher than the observed rate for white renters (one-tailed t-test; p < 0.001). The average eviction rate for Latinx renters (1.8 percent) was, however, significantly lower than the equivalent rate for white renters (one-tailed t-test; p < 0.001). These differences, although significant, were substantively minor compared with the black-white disparities. Figure 2 displays the distributions of filing rates (top panels) and eviction rates (bottom panels) for female and male renters by race/ethnicity. The average black female renter experienced an eviction filing rate of 6.4 percent, nearly twice that experienced by the average white female renter (3.4 percent). This disparity held for male renters as well, although the black-white gap among men was smaller (5.9 percent vs. 3.3 percent). 11 Average eviction rates for black renters were 3.5 percent for women and 3.3 percent for men. For white renters the equivalent rates were 2.0 percent for both women and men. The average female Latinx renter faced a 3.8 percent eviction filing rate and a 1.9 percent eviction rate; rates for their male counterparts were 3.4 percent and 1.7 percent, respectively.
We assessed the extent to which female renters were at disproportionate risk of eviction. Across all renters, the median ratio of female eviction rates to male eviction rates was 1.02, indicating that the risk of eviction was two percent higher for women Note: The columns labeled "Renters" refer to the share of the renter population that belonged to the given racial/ethnic group. The "Filings" columns provide the share of eviction filings in the county that were against members of that group. Standard errors are provided in parentheses.
than for men. Figure 3 plots the distributions of this ratio within each race/ethnicity category; the horizontal line at 1.0 represents gender equality in eviction rates. The median ratio was 1.04 among black renters and 1.09 among Latinx renters, meaning that eviction rates were four percent higher for black women than among black men and nine percent higher for Latinx women relative to Latinx men. By contrast, the gender disparity was below 1.0 among white renters (median of 0.97). Asian women were much less likely to be evicted than their male counterparts: the median ratio of rates was 0.82, indicating that Asian women were 18 percent less likely to be evicted than Asian men. These disparities amount to thousands of more evictions for women each year. Across 1,195 counties, we predicted that 341,756 female renters were evicted annually (95 percent confidence interval [CI] ± 485), approximately 15.9 percent more than the 294,908 evicted male renters (95 percent CI ± 458). The absolute and relative disparities in total evictions were greatest for black renters: 113,415 women evicted (95 percent CI ± 281) compared with 83,182 men (95 percent CI ± 259), or 36.3 percent more black women than black men evicted. For Latinx renters, we Figure 2: Distributions of eviction filing rates and eviction rates, by gender and race/ethnicity. Note: Data are presented as box and whisker plots, displaying distributions through five statistics: the median (the horizontal line within the white box for each state), the 25th and 75th percentiles (which form, respectively, the lower and upper bounds of each box), and two whiskers. Whiskers extend no more than 1.5 times the interquartile range (the distance between the first and third quartiles). Rates beyond the range of the whiskers are considered outliers and are plotted as individual points.
Black and Latinx renters who were filed against for eviction were most likely to be repeatedly filed against at the same address. Figure 4 displays the distributions of serial eviction filing rates by race/ethnicity. The average black renter experienced a serial eviction filing rate of 14.9 percent. On average, one in every seven black renters who was filed against for eviction was repeatedly filed against at the same address. The equivalent average rates were 13.2 percent for Latinx renters, 11.8 percent for Asian renters, and 9.8 percent for white renters.

Figure 3:
Distributions of female-male ratios of eviction rates by race/ethnicity. Note: Data are presented as box and whisker plots, displaying distributions through five statistics: the median (the horizontal line within the white box for each state), the 25th and 75th percentiles (which form, respectively, the lower and upper bounds of each box), and two whiskers. Whiskers extend no more than 1.5 times the interquartile range (the distance between the first and third quartiles). Rates beyond the range of the whiskers are considered outliers and are plotted as individual points.

Discussion
Drawing on data from 1,195 counties-covering more than one-third of all U.S. renter households-this study calculates eviction statistics by gender and race/ethnicity. Our analysis yielded three major findings. First, filing and eviction rates were, on average, significantly higher for black renters than for white renters. The share of eviction filings and eviction judgments against black renters was considerably higher than their share of the renter population. Second, black and Latinx female renters faced higher eviction rates than their male counterparts. Third, black and Latinx renters were most likely to be filed against serially for eviction. We discuss each of these findings in turn.
One in every five adult renters in our sample was black, yet one in every three eviction filings were served to a black renter. By contrast, whites made up more than Figure 4: Distributions of serial eviction filing rates by race/ethnicity. Note: Data are presented as box and whisker plots, displaying distributions through five statistics: the median (the horizontal line within the white box for each state), the 25th and 75th percentiles (which form, respectively, the lower and upper bounds of each box), and two whiskers. Whiskers extend no more than 1.5 times the interquartile range (the distance between the first and third quartiles). Rates beyond the range of the whiskers are considered outliers and are plotted as individual points.
half the population of adult renters (51.5 percent) but received only 42.7 percent of eviction filings. This resulted in a striking racial disparity. There were slightly fewer than 40 black renters for every 100 white renters in these counties. Yet for every 100 eviction filings to white renters, we estimated that there were nearly 80 eviction filings to black renters.
Because our results are unconditional, they may be explained in part by economic factors. Black households are more rent burdened and have higher levels of income volatility, compared with white households (Colburn and Allen 2018;Hardy and Ziliak 2014). They are also less likely to have access to resources that would help them weather unexpected events (Heflin and Pattillo 2006). It may also be the case that landlords and property owners employ differential treatment in the eviction process. If black tenants are not allowed as much leeway as their white peers when they fall behind on rent, they may be filed against more quickly and regularly. That explanation would be consistent with previous research indicating that the threshold for filing against white renters is higher than the threshold for filing against black and Latinx renters (Desmond and Gershenson 2017).
That black and Latinx female renters faced higher filing and eviction rates than their male counterparts confirms a finding identified by local studies (Desmond 2012;Desmond and Gershenson 2017;Desmond and Shollenberger 2015). Desmond's ethnographic work suggests two mechanisms that could explain these patterns. First, nonwhite women are more likely to be listed as leaseholders (and thus more likely to appear in the eviction records), owing to the fact that rates of unemployment and past incarceration are higher among their male counterparts. Second, children are a risk factor for eviction (Desmond et al. 2013), a dynamic that disproportionately affects single mothers.
Last, we found that black and Latinx renters were at greater risk of serial eviction filings than their white counterparts. To remain in place, tenants threatened with eviction must pay late fees and court costs in addition to settling rental debt. Leung and colleagues (2020) estimate that each eviction filing that does not result in housing loss costs renting households $180 in fines and fees on average, raising tenants' monthly housing cost by 20 percent. Racial disparities in serial eviction rates, then, have a real cash value and indicate that black and Latinx renters are disproportionately subjected to fines and fees through the eviction process. 12 Drawing on millions of court records, this study has produced evidence that black and Latinx renters in general, and women in particular, are disproportionately threatened with eviction and disproportionately evicted from their homes-and thus disproportionately exposed to the many documented negative consequences of eviction, from homelessness and material hardship to job loss and depression (Desmond and Kimbro 2015;Osypuk et al. 2012). Accordingly, sizable racial disparities in eviction rates documented here likely contribute to racial inequalities with respect to economic, social, and health outcomes.
The descriptive analyses presented here should motivate further research into racial/ethnic and gender disparities in eviction. State-and county-level eviction procedures shape how landlords and property managers use the courts (Leung et al. 2020). These policies may play a key role in producing the disparities documented for the first time here and may themselves be products of the demographic makeup of renter populations.
The Civil Rights Act of 1968, widely known as the Fair Housing Act, forbids practices that have a disparate impact on protected groups, including racial minorities and women, resulting in their denial of housing. The first step in making a disparate impact claim-the prima facie case-requires that a plaintiff identify a policy or practice to challenge, show a disparity in how this policy or practice affects a protected class, and establish a causal link between the policy/practice and the observed disparity (Schwemm and Bradford 2016:693). The data presented here may be especially helpful in demonstrating disparities (i.e., the second element of the prima facie case). They may also provide a roadmap for researchers and legal advocates attempting to identify legal practices and policies that are associated with particularly large disparities.
Notes 1 County-year data were considered reliable if the total number of LexisNexis filings in a county fell between 87 and 114 percent of the county courts' publicly reported total. For years when county court-level aggregates were not available, we extrapolated the most recently reported total a maximum of two years and applied the same validation range. We excluded county-years for which more than 60 percent of LexisNexis cases resulted in dismissals or had missing outcomes, suggestive of data quality problems. We excluded all county-years for which external validation was not possible. Averaging across countyyears, the mean county in our sample had a coverage percentage of 97 percent (standard deviation of 5.03 percent) relative to court-reported statistics. Eviction rates may be slightly under-or overestimated in any specific county, but the population-level effect is likely to be a very slight underestimate of rates. Notably, under-or overreporting of evictions in a county-year is likely random and does not disproportionately affect members of one gender or racial/ethnic group. This minimizes the risk of skewed cross-race or cross-gender comparisons.
in findings across studies is likely due to the unconditional nature of estimates presented here, as well as the different units of analysis (individual vs. Census tract). Interactions between individual and neighborhood factors in predicting serial eviction filing merit further analysis.