Deporting the American Dream : Immigration Enforcement and Latino Foreclosures

Over the past decade, Latinos have been buffeted by two major forces: a record number of immigrant deportations and the housing foreclosure crisis. Yet, prior work has not assessed the link between the two. We hypothesize that deportations exacerbate rates of foreclosure among Latinos by removing income earners from owner-occupied households. We employ a quasi-experimental approach that leverages variation in county applications for 287(g) immigration enforcement agreements with Immigration and Customs Enforcement and data on foreclosure filings from 2005–2012. These models uncover a substantial association of enforcement with Hispanic foreclosure rates. The association is stronger in counties with more immigrant detentions and a larger share of undocumented persons in owner-occupied homes. The results imply that local immigration enforcement plays an important role in understanding why Latinos experienced foreclosures most often. The reduced home ownership and wealth that result illustrate how legal status and deportation perpetuate the racial stratification of Latinos.

O VER the past decade, Latinos in the United States have been affected by two major crises, neither of which has fully ended.First, immigrant deportations reached record levels-rising from 165,168 in 2002 to 319,382 in 2007 and 438,421 by 2013, or 3.6 million deportations in total (Gonzalez-Barrera and Krogstad 2014; U.S. Department of Homeland Security 2014).Deported immigrants have been almost exclusively Latino: 94 percent of those deported since 2005 have been Mexican or Central American despite their two-thirds share of all unauthorized persons (Passel and Cohn 2014;U.S. Department of Homeland Security 2014).The impacts of deportations are not limited solely to the unauthorized because millions live in "mixed status" households along with documented persons (Taylor et al. 2011).In the second crisis, Latino households lost their homes to foreclosure more often than any other ethno-racial group in the recent housing crash (Bocian et al. 2011;Hall, Crowder, and Spring 2015b;Rugh 2015a).Yet, while millions of Latino households were devastated by the crises of record deportations and foreclosures, we are unaware of prior work that connects the two.
In this article, we aim to establish an empirical link between Latino deportations and Latino foreclosures.We hypothesize that the surge in Latino deportations in the mid-2000s that preceded the foreclosure crisis exacerbated rates of foreclosure among Latino homeowners.Because Latino immigrants often live in mixed legal status, owner-occupied households, we expect the deportation of a wage-earning adult in the household to raise the risk of home foreclosure.Latino household foreclosure rates should be uniquely affected because 85 percent of deportees are estimated to be male immigrants from Latin America previously in the labor force To identify these effects, we leverage geographic and temporal variation in application and approval of local 287(g) immigration enforcement agreements with Immigration and Customs Enforcement (hereafter ICE).We examine how the deportation of Latino immigrants may be causally related to Latino foreclosure rates by taking advantage of variation in the outcomes of 287(g) applications and employing inverse probability of treatment weights to a difference in difference framework (see Kostandini, Mykerezi, and Escalante 2014;Potochnick, Chen, and Perreira 2016).

Background
In the post-9/11 era, increased immigration enforcement in the U.S. interior has led to a record number of deportations of immigrants already living here (i.e., removals, not returns of recent border crossers).Total interior removals increased steadily from 122,587 in 2002 to 207,776 in 2006, then rose further to 300,135 in 2009 (2.4 times the 2002 total) and remained elevated at 282,156 in 2012 (Immigration andCustoms Enforcement 2013). 1 This intensification of interior enforcement efforts coincided with an unprecedented dispersal of Latino immigrants across the United States.Until 1990, for instance, around nine in 10 recent Mexican immigrants settled in California, Illinois, and Texas, a share that dropped to five in 10 by 2005 (Massey and Capoferro 2008).Ironically, increased United States-Mexico border enforcement in the 1990s has been causally linked to Latino immigrant dispersal, which appears to have partly precipitated the ensuing interior enforcement.(Bohn and Pugatch 2015;Massey, Durand, and Pren 2016).
The surge in immigrant detentions also increased in lockstep with Latino household foreclosure rates, which also rose to record levels (Figure 1).According to 2013 data from the Panel Survey of Income Dynamics, cumulative foreclosure rates since 2001 stood at 19 percent for Latinos, 15 percent for blacks, and 8 percent for whites (Hall et al. 2015b).In the early years of the U.S. housing crash, blacks and Latinos appeared to be equally devastated by foreclosures because of factors such as discriminatory subprime lending (Rugh and Massey 2010).
However, the discrepancy between black and Latino foreclosure rates that emerged and widened over time has spurred inquiry into the role of nativity and legal status in explaining the Latino disadvantage (Allen 2011;Kochhar, Gonzalez-Barrera, and Dockterman 2009;Pfeiffer et al. 2014;Rugh 2015a;Rugh 2015b;Rugh and Allen 2015).For example, Cuban immigrants (who enjoy immigration protections) and Puerto Ricans (who are U.S. citizens) were not as negatively affected as Mexican Americans by the housing crisis (Cahill and Franklin 2013;Kuebler and Rugh 2013).

Section 287(g) Local Immigration Enforcement
The spike in interior immigration enforcement has largely been a result of changes in policies dictating the role of local governments in facilitating detention.In The 287(g) immigration enforcement agreements provided local officials unprecedented latitude to pursue, detain, and deport immigrants already living in the United States.In the late 1990s and early 2000s, section 287(g) was adopted as a very focused program on criminal removals, but with the rapid growth of Latinos in new destinations (particularly in the southeast), 287(g) became viewed as an effective local tool to remove all unauthorized persons (see Capps et al. 2011).Consequently, the proliferation of the agreements happened in mid-2000s and the counties that signed agreements during this time were home to more than 5 million Central Americans and Mexicans and an estimated 3 million undocumented immigrants, or one-quarter of the total in the United States at the time (authors' calculations).
Compared to the more widely studied Secure Communities initiative (e.g., Miles and Cox 2014;Watson 2013), the 287(g) program data better suit our purposes for three reasons.First, although Secure Communities was first piloted in late 2008 in a handful of counties, it was mainly implemented from 2010 to 2013, too late to cause foreclosures in 2006 to 2010.Secure Communities was more focused on criminal removals because it followed 287(g).Its policies, while still criticized on civil rights grounds, were less controversial compared to the allegations of racial profiling against officials carrying out 287(g) enforcement (e.g., Capps et al. 2011;Lacayo 2010).Second, the rollout of 287(g) was uneven across counties in various states without relation to the United States-Mexico border.In the case of Secure Communities, time and distance from the border were correlated: it was rolled out first in border regions and then slowly into the rest of the interior United States (Miles and Cox 2014).
The final reason we select 287(g) local enforcement is that it drove a historic increase in noncriminal deportations (Capps et al. 2011;Watson 2013).In the absence of 287(g) enforcement, many otherwise law-abiding Latino immigrants may not have been deported (Golash-Boza 2015).The noncriminal share of deportations for traffic offenses or no offense ranged tremendously across locations and tended to be highest in counties undergoing rapid Latino population growth (Capps et al. 2011).Thus, 287(g) data provide a better proxy for immigrant detentions that otherwise would not have occurred in the context of the criminal justice system.
To the extent that 287(g) enforcement and related initiatives increasingly removed longtime residents for noncriminal offenses in established undocumented and mixed status families, they would be more likely to be members of homeowner households.In this regard, 287(g) noncriminal removals come closer to identifying the variation in immigration deportations most likely to affect owner-occupied Latino households.Based on the prior literature and the contours of 287(g) agreements, we hypothesize that local 287(g) enforcement will lead to more foreclosures in counties with higher rates of home ownership among households with Latino immigrants (greater exposure to the risk of foreclosure).The next section details the conceptual model for this hypothesis.
Linking Deportation, Home Ownership, and Foreclosure Figure 2 presents our conceptual model of pathways from deportation to foreclosure.Latino deportations and Latino out-migration in response to increased local immigration enforcement decrease the number of immigrant adult wage earners in owner-occupied Latino households.Many households pool resources across families and mixed legal statuses to achieve and sustain homeownership (Golash-Boza 2015;Guelespe 2013;Schueths 2015).Some argue that immigrants are drawn to homeownership because of its symbolism in the "American Dream" narrative and are willing to enter homeownership by stretching their earning power (Thomas 2013). 2  Latino immigrants put down roots in the United States, including household home ownership across mixed legal statuses.Among those deported, the median length of U.S. residence is 14 years (Brabeck and Xu 2010).In studies of mixed legal status couples, home ownership ranged from 40 percent to 50 percent, and the presence of extended family was common (Guelespe 2013;Schueths 2015).Legal permanent residents are more likely to be homeowners than the undocumented and are also deported via local police-based enforcement (Golash-Boza 2015; Hall and Greenman 2013).In the 42 counties with 287(g) agreements analyzed in this article, 72 percent of the 1.2 million households with at least one unauthorized adult Moving across our conceptual model in Figure 2, we hypothesize that the ensuing loss of income formerly devoted to mortgage payments raises the likelihood of household foreclosure.When deportations of Latino immigrants began to accelerate in 2006 to 2008, wage-earning adults were removed from these Latino homeowner households, which subsequently contributed to an elevated rate of foreclosures in 2008 to 2011.Most deported immigrants previously contributed to household mortgage payments, even though they were not likely cosigners on actual home loans (Guelespe 2013;Rugh and Allen 2015;Zayas 2015). 3More than 90 percent of those deported wage earners are men; their labor market participation rates and incomes tend to exceed those of the women left behind, who are often primary caretakers of dependent children (Golash-Bozaa and Pierrette Hondagneu-Sotelo 2013; Rosenblum and McCabe 2014).
Prior research strongly suggests that household income loss plays a role in our conceptual chain between immigration enforcement and foreclosure.The consensus of most research estimates postdeportation losses in household income to be around 50 percent (Capps et al. 2015;Chaudry et al. 2010;Koball et al. 2015;Brabeck and Xu 2010).In addition, expenditures on legal fees and immigration bonds drained the finances of many immigrant families facing deportation, exacerbating income losses (Koball et al. 2015).Evidence of reductions in migrant remittances also suggests that enforcement reduces the incomes of households with deported immigrants.Undocumented Mexican migrants residing in counties that adopted 287(g) agreements reduced their remittances by 35 percent postadoption compared to those in other counties; in contrast, remittances did not decline among documented migrants (Amuedo-Dorantes and Puttitanun 2014).
At the end of the chain of risk factors in Figure 2, home foreclosure is the final consequence.Qualitative and quantitative findings point to reduced home ownership as the ultimate outcome of immigration enforcement and household income loss. 4Among a sample of Latino families that owned their homes and experienced an immigration arrest, incomes losses were severe and half of them lost their homes within one year (Chaudry et al. 2010). 5Because the time from detention to deportation, mortgage default, and subsequent foreclosure usually spans more than one year, in our model we expect the impact of enforcement to last longer, perhaps several years.In summary, we find evidence for the linkages in our conceptual model in the prior literature on mixed status households and couples, immigrant home ownership, and the effects of detention and deportation on incomes.

Data and Methods
To evaluate the impact of deportations on foreclosures, we compile data from a variety of sources to assemble a panel file of relevant county-level information.Our measures of county foreclosures are based on the population of foreclosure filings as collected by RealtyTrac.Specifically, we obtained the complete database of all foreclosure-related documents collected in virtually every county in the country by RealtyTrac between 2005 and 2012.The full database of more than 20 million foreclosure documents was reduced to "first-visible" foreclosures, which refers to the point when a foreclosed residential property is either auctioned or repossessed (see Hall, Crowder, and Spring 2015a for additional information on the algorithms used during this process).In total, this procedure estimates that 9.5 million homes were foreclosed between 2005 and 2012 (see Hall et al. 2015b).
The RealtyTrac foreclosure filings include detailed information on the location and timing of foreclosure but lack information on the ethnoracial identity of homeowners.In order to estimate the number of foreclosures for major racial/ethnic groups, we use Summary File 1 of Census 2000 to assign a probability weight to each foreclosure based on the racial composition of the census block in which the housing unit is located.For example, a foreclosure occurring in a block that was 70 percent white and 30 percent black would be classified as 0.7 white and 0.3 black. 6 The same procedure is used to delineate the number of housing units occupied by each racial group in every census block in 2000.We use these imputed counts of foreclosures and housing units to calculate race-specific foreclosure rates (foreclosures per 100 housing units) annually for each year from 2005 to 2012.Among the 433 counties that form our main analytic sample, the average annual foreclosure rate is 1.05 percent, with average annual rates for whites (0.96 percent) being lower than rates for blacks (1.10 percent) and Hispanics (1.01 percent). 7As expected, there is substantial variation over time in these rates.Average annual foreclosure rates for Hispanics, for example, were 0.24 percent in 2005 but 1.62 percent in 2010.Moreover, variation across counties is considerable: at the peak of the crisis, for instance, Hispanic foreclosure rates were 11.6 percent in Maricopa, Arizona, but 0.02 percent in Wichita, Texas.
The focal independent variable in our analysis is whether a county has implemented a 287(g) agreement with the Department of Homeland Security.The full set of historical and active agreements was released by ICE via the Freedom of Information Act (FOIA) (U.S.Department of Homeland Security 2016) and contains the date of implementation and type of agreement (jail or task force).The majority of these agreements were made by county sheriff's offices, although several were to municipalities within counties, which we apply to the entire county. 8We also obtained information through a separate FOIA request on the number of persons identified for removal by ICE as well as those who were detained.In FY 2010, for example, 47,252 individuals were identified for removal and 80.2 percent of these individuals were ultimately detained by ICE.We use these counts to describe the intensity of the impact of enforcement by dividing the sum of removals in each county by the total foreign-born population of the county.This approximation of the share of immigrants detained by 287(g) is highly variable, with Alamance, North Carolina, having identified 24 percent of its immigrant population for removal while Fairfield, Connecticut, identified 0 percent.Lastly, we obtained information via the FOIA (U.S.Department of Homeland Security 2015) on the counties (or jurisdictions within counties) that submitted an application to enter into a 287(g) agreement but were rejected by the Department of Homeland Security.Detailed information on the reason for denial were not included in the FOIA response, but media reports and correspondence between agencies indicates that rejections were a result of a county not having the resources to staff immigration-related intakes or having the capacity to hold those identified for removal.
To account for differences across counties in immigration enforcement, we use propensity scores methods (described below) to predict 287(g) participation as a function of a set of covariates expected to be related to local enforcement.To capture demographic variation across counties, we include measures of the percent of the population that is Hispanic, the percent of Hispanics of Caribbean descent, the percent foreign-born, and total population size (logged), all of which are derived from Census 2000.Median household income (from Census 2000) is included to account for economic differences.To capture differences in housing markets and to consider the possibility that immigration enforcement was more intense in markets with rapid housing cost growth, we include measures of the percent of the population that lived in owner-occupied housing, 2000-2005 housing value growth rates from Zillow.com, the percent of home loans underwritten with no documentation, the percent of loans with high-cost rate spreads, and the average backend debt-to-income ratio among borrowers (see Rugh 2015a).Indicators for census region and metropolitan status are also included.
A central variable in the treatment model and in moderation analyses is the number of unauthorized immigrants in a county.To estimate county-level unauthorized populations, we employ the data fusion approach developed in Capps et al. (2013) that applies correlations between legal status and other characteristics observed in the 2001 Survey of Income and Program Participation to public-use microdata samples (PUMS) from Census 2000 using multiple imputation (see Bachmeier, Van Hook, and Bean 2014;Van Hook et al. 2015).This process is used to generate two relevant estimates: the number of unauthorized persons and the percent of unauthorized immigrants living in owner-occupied housing.This reliance on PUMS data to complete this process, combined with the exclusion of three additional counties (Rockingham, New Hampshire, Strafford, New Hampshire, and Douglas, Oregon) with incomplete foreclosure data, restricts our main analytic sample to 433 counties with boundaries that closely follow census PUMAs. 9

Analytic Approach
Our analysis leverages variation in the timing and location of 287(g) programs to estimate the impact of immigration enforcement on home foreclosures.In the first stage of our analysis, we determine the propensity score that a county implements a 287(g) program as the conditional probability of assignment to the treatment, given a set of demographic, economic, and housing covariates.The probit is expressed as: where T is an indicator of whether a county implemented 287(g) and X represents the covariates discussed above, including total and foreign-born populations, unauthorized shares, racial composition, median incomes, ownership rates, risky lending measures, housing price growth, census region, and metropolitan status.We estimate this model separately for the full analytic sample of 433 counties and the smaller quasi-experimental set of counties that applied for 287(g) (including the rejected applicants).
We use the propensity scores from these models to generate weights equal to the inverse of the probability of treatment (IPT): This gives greatest weight to treated (287[g]) counties with the lowest probability of treatment and untreated counties with highest probability of treatment (see Morgan and Todd 2008).As discussed below, this model is successful in balancing covariates between treated and control groups and more so in the quasi-experimental sample than in the larger set of counties.
The main part of our analysis assesses the impact of immigration enforcement on foreclosures using a doubly-robust difference in difference model with IPT weighting.The basic model is expressed as follows: where f oreclosures kct is the foreclosure rate for racial/ethnic group k in county c and year t; τ ct is a set of event time dummies, defined in two-year intervals such that it includes the two periods (up to four years) before implementation of 287(g), the implementation year, and the three periods (up to six years) following implementation.State fixed effects are represented by α s , δ t are year effects, and λ s t are state-specific linear time trends; X c is the set of covariates in the treatment model.These estimates incorporate the IPT weights described above; robust standard errors are clustered at the county-level.Our main interest lies in the estimates of τ ct , which indicate how foreclosure levels differed between counties with and without 287(g) before and after implementation of the program.

Results
Participation in the 287(g) program, and immigration enforcement more broadly, is variable across time and space.These two sources of variation are shown in the two panels of Figure 3, which identify the location of all 287(g) participants (panel A) and the timing of their implementation (panel B).More specifically, the map identifies all counties that applied for 287(g) between 2005 and 2012, with green shades identifying counties that were approved for participation and red shades representing counties with applications that were rejected.The map shows considerable variable across the country in 287(g) involvement, but noticeable clusters

Predicting 287(g) Participation
The first stage in our analysis predicts the likelihood that a county will implement a 287(g) agreement.The results of this treatment model are summarized in Table 1.Specifically, Table 1 shows average marginal effects from the probit model of 287(g) participation for our two analytic samples of counties: the broad set of counties with valid information on key covariates (N = 433) and the subset of counties that applied for 287(g) (N = 104).
The model estimates indicate that only three county characteristics are significantly related to 287(g) implementation: region, total population, and the percent of Hispanic immigrants estimated to be unauthorized.The estimates suggest that larger southern counties with high unauthorized shares were most likely to adopt a 287(g) agreement.The estimates for the smaller set of counties that applied for 287(g) are similar in magnitude but are less precise.Specifically, the coefficient estimates on unauthorized shares, total population, and region suggest a tendency for counties with accepted agreements to have larger unauthorized shares and to be located in the southern or western regions, but none of these coefficients are significantly different from zero.
We employ IPT propensity weighting to balance covariates between control and treatment counties.In the larger set of counties, the overidentification test for balancing has a χ 2 = 0.94 (p < 1.00); for the set of 287(g) applicant counties χ 2 = 4.50 (p = 0.99).Weighted means for individual covariates in the treatment model are shown in Table A1 of the online supplement.As shown there, a significant difference between treated and control counties in unauthorized shares exists even with the IPT weighting.However, in the sample of 287(g) applicant counties, balancing of all individual covariates is achieved.

The Impact of 287(g) on Foreclosures
Results from models evaluating the impact of 287(g) adoption on county foreclosures are shown in Table 2.The table is organized to show the treatment effects across event time, and models are estimated separately for overall county foreclosure rates and rates for Hispanics.All models incorporate IPT weights and include state and year fixed effects, state-specific trends, and the treatment model controls.
The results in Table 2 show clearly that 287(g) adoption was associated with an increase in county foreclosures.Specifically, the estimates indicate that foreclosure rates in 287(g) counties did not differ from non-287(g) counties in the years preceding implementation of the program-i.e., the coefficients on the pre-287(g) terms are very small and statistically indistinguishable from zero.The coefficients in the years following adoption, however, indicate that foreclosure rates are substantially (and significantly) higher in 287(g) counties.The estimate of 0.6356 in the two years after implementation indicates that foreclosure rates in 287(g) counties were about 0.64 percentage points higher than in non-287(g) counties.This effect is equivalent to 0.42 standard deviations from the mean foreclosure rate.There is no strong evidence that the 287(g) effect wanes over the first two postimplementation periods, as the confidence intervals on the post-287(g) coefficients are all overlapping.As expected, the results for Hispanics are stronger, with our models suggesting that 287(g) was associated with a sizeable increase in rates of Hispanic foreclosure in all three postimplementation periods.
Table 2 provides evidence that county immigration enforcement augmented local foreclosures.Yet, the estimates from Table 2 are potentially confounded by characteristics of counties that select them into 287(g) participation.While the IPT weights help to adjust for observable differences between 287(g) and other counties, unobservable features of counties are likely to bias our estimates.Moreover, the propensity scores are not completely successful in this full sample to balance on all key characteristics-most importantly, the share of Hispanics who are unauthorized.To partly resolve these issues, our preferred analysis is based on the set of counties that applied for 287(g), including both the rejected and accepted applications.
Results from models for the 104 counties that applied for 287(g) are summarized in Table 3.The estimates in Table 3 provide compelling evidence that 287(g) deepened the housing crisis, with overall foreclosure rates being about 0.4 percentage points (or 0.20 standard deviations) higher in counties that adopted a 287(g) program relative to those that applied but did not adopt the program.The estimates are somewhat imprecise, with only the effect for one to two years following adoption reaching significance at the 10 percent level, but the pattern of difference is strongly consistent with the expectation that 287(g) intensified foreclosure rates.
Hispanic foreclosures than other groups.The posttreatment estimates for black foreclosures are all positive but comparatively small and not statistically significant.
The estimates for whites, however, indicate that 287(g) may have had broader impacts on foreclosures.As our race-specific foreclosures are constructed by weighting each foreclosure by the racial composition of its census block, we are potentially misclassifying "white" and "Hispanic" foreclosures if Hispanics living in integrated or whiter neighborhoods face high risks of foreclosure.To the extent that this is true, our estimates for Hispanic foreclosures will likely be biased downward and the estimates for white foreclosures pushed upward.To test this possibility, we calculated Hispanic-white segregation for all counties in our analysis and estimated models separately for counties with low Hispanic-white segregation (dissimilarity <0.6) and those with high segregation (dissimilarity ≥0.6). 10 These estimates are shown in Table A2 of the online supplement and demonstrate clearly that the effects of 287(g) on white foreclosures were limited to those areas where whites and Hispanics were more likely to coreside.

Exploring Mechanisms of the Impacts
This analysis suggests that immigration enforcement increased home foreclosures among Hispanic households.In the final stage of our analysis, we consider the moderating roles of enforcement intensity and the presence of unauthorized persons in owner-occupied housing as a means to explore possible pathways connecting 287(g) to housing foreclosure.Specifically, in the first two columns of Table 4, we stratify our sample of 287(g) counties into those that detained a large share (>5 percent) of the local immigrant population and those that detained relatively small shares (<5 percent).
The results clearly show that the impact of 287(g) on Hispanic foreclosures was especially strong in counties with high detention rates (with strong treatment intensity) but smaller and statistically insignificant with 287(g) counties with lower detention rates (with weaker intensity).These relationships are shown in graphical form in Figure 4, where the red circles represent point estimates for weakly enforced counties, and the blue Xs represent strongly enforced counties (attached to 90 percent confidence intervals).
Our conceptual model of the link between deportations and foreclosures views the loss of adult wage earners in homes with mortgages as one of the central mechanisms.Following this logic, the impacts of 287(g) on foreclosures should be heightened in counties where a larger portion of the unauthorized live in owneroccupied homes.The third and fourth columns of Table 4 present results that are consistent with this argument.Specifically, the post-287(g) effect estimates are substantially higher in counties where the unauthorized have high ownership occupancy (more than 30 percent reside in owner-occupied housing) than in areas where the unauthorized are less likely to live in owned homes (<30 percent).Again, we illustrate this difference graphically in Figure 5, which clearly shows the large and significant point estimates of 287(g) on foreclosures in the years following adoption for counties where unauthorized persons are more likely to live in owned homes.

Discussion
In this article, we took advantage of county variation in the timing and geographic location of 287(g) applications to model difference in difference estimates (combined with inverse probability of treatment weights) of the effect of local immigration enforcement on rates of Hispanic foreclosure between 2005 and 2012.Our results document a clear pattern in counties that implemented a 287(g) agreement experienced a surge in Latino foreclosures.More specifically, our preferred models estimate that in the two years following adoption, Latino foreclosure rates in 287(g) counties were about 0.68 percentage points higher than in non-287(g) counties (or 0.39 standard deviations larger than the mean).Consistent with our expectations, the magnitude of this association is nearly twice as large in counties where more immigrants are identified for deportation and where a higher share of the undocumented population resides in owner-occupied homes.
While our analysis is the first that we are aware of to propose and detect an effect of immigrant deportations on Latino foreclosures, our analysis does face potential limitations.Our quasi-experimental estimates are not as precise because of the limited statistical power that stems from the sample size.We note that our ecological data using census block racial shares are indirect estimates of foreclosures among Latino households; at the same time, the data represent the most comprehensive database on foreclosures that we are aware of.Future analyses should explore the use of individual data nested within local immigration enforcement districts to gain greater leverage on the proposed mechanisms and overcome these power issues.We caution that we cannot interpret our estimates to be strictly causal.However, the results may be interpreted as reasonably close to causal if we assume that counties with accepted and rejected 287(g) applications are similar in terms of unobservable characteristics that predict foreclosure.This assumption seems reasonable in light of previous scholarship that employs similar strategies of statistical inference to infer causality.Finally, we note that our conceptual model is not intended to portray potential effects on non-Latino households, local businesses that cater to Latino immigrants (Hagan, Rodriguez, and Castro 2011), or the ultimate impact on Latino household wealth and opportunity (Keister, Vallejo, and Borelli 2015).We view this latter limitation, however, as a favorable invitation for further research on the compounding effects of these processes.
Our analysis has important implications for housing and immigration policy as well as theoretical and empirical work on immigrant incorporation and Latino residential stratification.During the housing crash and ensuing Great Recession, increased local immigration enforcement that removed Latino immigrants from homeowner households worked at cross-purposes with federal initiatives to help troubled homeowners stay in their homes (see Immergluck 2015).As Menjívar (2014) recently concluded, interior immigration enforcement has "far-reaching and unintended consequences" (p.357).By lowering incomes, home ownership, and immigrant demand for housing in owner and rental markets, deportations may partly explain the relatively prolonged slump in housing markets in metropolitan regions with high levels of enforcement like Phoenix and Atlanta.Future work should investigate the possible impact of immigration enforcement on housing policies and outcomes that affect the broader population.
By joining the research on immigration enforcement and home ownership, our analysis also contributes to the theoretical and empirical scholarship of immigrant incorporation (Alba and Nee 2005;Bean, Brown, and Bachmeier 2015;Myers 2007;Park and Myers 2010;Telles and Ortiz 2008;Vallejo 2012), illegality (Chávez 2008;De Genova 2002;Dreby 2015;Massey, Durand, and Pren 2014;Menjívar 2006), mixed status households (Capps and Fortuny 2006;Capps et al. 2015;Cardoso et al. 2014;Dreby 2012), and the residential stratification of Latinos (Fischer and Tienda 2006;Hall 2013;Hall and Stringfield 2014;Rugh 2015a).Our conceptual model and analysis results strongly suggest that unauthorized status leads to a chain of events from deportation, lost household income, foreclosure, and, ultimately, lost wealth.The results thus lend support to the delayed assimilation model (Brown 2007;Bean et al. 2015) but with a slight twist.What we have analyzed here does not appear to be as straightforward as delayed or downward assimilation: Latino homeowners who experience deportation of a household member already achieved a measure of upward mobility-home ownership.In this regard, our analysis squares with others who find high rates of intergenerational mobility (Vallejo 2012;Kiester, Vallejo, and Borelli 2015).Yet, these intergenerational gains in home ownership may be vulnerable to foreclosure during an era of mass deportation of Latino immigrants.
The results of this analysis underscore the emergence of legal status as an axis of stratification that uniquely disadvantages Latinos through racially disproportionate rates of deportation.The ensuing consequences of foreclosure for Latino households of reduced home ownership and wealth illustrate how legal status and racialized patterns of deportation act to stratify Latinos' housing outcomes and impede social mobility.Our results raise important questions about the fragility of social mobility achieved via home ownership and Latino stratification along America's evolving racial color line (e.g., Bonilla-Silva 2002;Gans 2005;Golash-Boza 2006;Lee and Bean 2010;Lichter 2013;Massey 2013).Rapid ascent into government-subsidized home ownership was a sufficient condition for wealth-building among whites in the twentieth century.For millions of Latino households, plummeting home equity and home ownership have been devastating in the twenty-first century context of risky lending and the foreclosure crisis.
In the continuing deportation crisis, undocumented and mixed legal statuses are also linked to the residential and racial stratification of Latinos today.As the country prepares for the possibility of increased enforcement and mass deportations, the lessons learned from this analysis take on new meaning and have increased relevance.Policymakers and advocates would be wise to consider the wide-reaching repercussions of enforcement for individuals, families, and communities, both in the short and long terms.3 Based on recent examination of thousands of Latino loan borrowers and the types of identity documents used to sign mortgages (e.g., driver's license, passport, green card, etc.), Rugh and Allen (2015) estimate that at most five to eight percent of Latino borrowers who signed mortgage loans were recent noncitizen residents or unauthorized immigrants without driver's licenses.
sociological science | www.sociologicalscience.com 4 Longitudinal data on Hispanic and immigrant movers further underscore how nativity and citizenship status combined to undermine home ownership among Latino households.Net of other factors, immigrant Hispanics are no less likely than native Hispanics to transition to first-time home ownership, with transitions primarily during the 2004-2006 housing boom; however, immigrant Hispanics were 1.7 times more likely to exit ownership, mainly during the 2007-2009 housing crash (Mundra 2013).Among immigrants of all races and ethnicities, noncitizens were roughly 1.5 times more likely to exit ownership (Mundra 2013).These longitudinal trends suggest that first-time ownership among Latino noncitizen households was especially vulnerable during the housing downturn.6 The implicit assumption in this approach is that racial group members face the same risk of foreclosure within neighborhoods and are only differentiated by the relative size of their group.While most of the variation in foreclosures exists across neighborhoods, there is some evidence that black and Hispanic mortgage holders are more likely than whites in the same neighborhood to foreclose.The implication is that we are likely slightly underestimating foreclosures to blacks and Hispanics.Importantly, however, there is no reason to believe this source of measurement error varies by county 287(g) status.
influence of prison populations on these measures, we exclude from these calculations blocks where 25 percent or more of the population is institutionalized.

Figure 2 :
Figure 2: Conceptual model of Latino deportations and Latino household foreclosures.

Figure 4 :
Figure 4: Effects of 287(g) on Hispanic foreclosure rates, by enforcement intensity.

Figure 5 :
Figure 5: Effects of 287(g) on Hispanic foreclosure rates, by % of undocumented in owner-occupied housing.

Notes1
The surge in noncriminal deportations was even steeper.Noncriminal removals doubled from50,901 in 2002 to 115,513 by 2006, peaked at 173,212 in 2009 (3.4  times the 2002 total), and remained elevated at145,802 in 2012145,802 in   (ICE 2013)).Warren and Warren (2013) report that the number of unauthorized immigrants removed after living in the United States for more than six months doubled from65,279 in 2000 to 133,190 in 2007, then rose  to 164,839 in 2009.Extrapolating the prior rate of increase (5 percent) in deportations in the years leading up to 2002, one would expect 1.62 million deportations from 2003-2012.Instead, the actual total was a staggering 2.43 million.In other words, as a result of the post-2002 surge in deportations, there were an estimated additional 810,000 immigrants deported-and 72 percent were noncriminal deportations.2During the housing boom, multiple families were more common among Latino owneroccupied (homeowner) households, an indicator that Latinos were more likely to rely on multiple wage earners to make mortgage payments.In 2006, 15.2 percent of Hispanicheaded homeowner households in the United States contained multiple families, nearly twice the rate (7.8 percent) among non-Hispanic white homeowner households.In the expensive housing markets of Los Angeles and Washington, D.C., the rate among Hispanic homeowner households stood at 21.7 percent and 25.4 percent, respectively, compared to only 10.6 percent to 8.2 percent of such white households (authors' calculations based on March 2006 Current Population Survey).

5
Because the interviews took place in 2008 and 2009, 50 percent is likely a conservative estimate (right-censored) because other families may have lost homes later.Nonetheless, results from Chaudry et al. (2010) are a useful indicator of what the immediate first-year effect of detention and deportation may have on home ownership stability.

Table 1 :
Average marginal effects from probit of 287(g) participation.
p < 0.10; standard errors clustered at county level.