Federal-Local Partnerships on Immigration Law Enforcement: Are the Policies Effective in Reducing Violent Victimization?

Abstract

Research Summary

Our understanding of how immigration enforcement impacts crime has been informed by data from the police crime statistics. This study complements existing research by using longitudinal multilevel data from the National Crime Victimization Survey (NCVS) for 2005–2014 to simultaneously assess the impact of the three predominant immigration policies that have been implemented in local communities. The results indicate that the activation of Secure Communities and 287(g) task force agreements significantly increased violent victimization risk among Latinos, whereas they showed no evident impact on victimization risk among non-Latino Whites and Blacks. The activation of 287(g) jail enforcement agreements and anti-detainer policies had no significant impact on violent victimization risk during the period.

Policy Implications

Contrary to their stated purpose of enhancing public safety, our results show that the Secure Communities program and 287(g) task force agreements did not reduce crime, but instead eroded security in American communities by increasing the likelihood that Latinos experienced violent victimization. These results support the Federal government’s ending of 287(g) task force agreements and its more recent move to end the Secure Communities program. Additionally, the results of our study add to the evidence challenging claims that anti-detainer policies pose a threat to violence risk.

The enforcement of immigration laws and its effect on crime are highly debated in the United States (Koper, Guterbock, Woods, Taylor, and Carter, 2013; Kubrin, 2014; Treyger, Chalfin, and Loeffler, 2014; Martinez and Iwama, 2014; Rosenfeld, 2014). For more than a century since the American Civil War, the federal government held near-exclusive power to regulate immigration (Chy Lung v. Freeman, 92 U.S. 275 [1876]). Even though there was increased pressure on state and local officials to become involved in immigration enforcement in the 1980s and 1990s (Stumpf, 2006), the Department of Justice (DOJ)’s Office of Legal Counsel (OLC) stated in a memorandum in 1996 that state and local police held no authority to enforce civil immigration violations (OLC, 1996). In 2002, however, the federal guidelines changed to request (and at times demand) local help in the enforcement of immigration laws after the terrorist attacks of September 11, 2001 drew heavy criticism of the lack of cooperation between federal and local law enforcement (Arnold, 2007; Lasch, 2013). A new DOJ memorandum in 2002 announced that local officials have inherent authority, subject to federal preemption, to make arrests for federal immigration violations (OLC, 2002). Since then, there has been a large expansion of state and local involvement in immigration enforcement (Amuedo-Dorantes and Lopez, 2022; Kandel, 2016).

The participation of local officials in the enforcement of immigration laws poses challenging issues for policymakers as well as for researchers involved in the evaluation of public policies. Under the new federal guidelines, the U.S. Immigration and Customs Enforcement (ICE) agency, a division of the Department of Homeland Security (DHS) created in 2003, has actively pursued partnerships with local officials to enforce immigration laws through, most notably, the Section 287(g) and Secure Communities programs (Rosenblum and Kandel, 2012). These programs were designed to remove serious criminal aliens from the United States (American Immigration Council, 2014; Golash-Boza and Hondagneu-Sotelo, 2013), but they have created many controversies and debates about the legality and impact of local involvement in immigration law enforcement, often because the vast majority of individuals affected by the programs are not serious criminals (Coleman, 2012; Kubrin, 2014; Lacayo, 2010; Martinez and Iwama, 2014; TRAC, 2014a; Zatz and Smith, 2012). A substantial number of local jurisdictions have optioned, amid the debates, to decline cooperation with federal authorities or have enacted immigrant protective anti-detainer or so-called “sanctuary” policies to limit the role of local officials in immigration enforcement (Graber and Marquez, 2016; Rice, 2017; Ridgley, 2013).

In managing the varied immigration enforcement approaches, the federal and local governments have spent significant amounts of resources and taxpayer money. Since its inception, the overall budget of ICE has increased from $3.3 billion in 2003 to $8.3 billion in 2021 (American Immigration Council, 2021a). The appropriations for the Secure Communities program alone totaled $1.1 billion from 2008–2014 (Cantor, Noferi, and Martinez, 2015), prior to the reactivation of the program in 2017. Though smaller in scope, the 287(g) program has been estimated to cost Americans an additional $500 million in federal funding since 2006 (American Immigration Council, 2021b). A major rationale for these investments is that policies such as 287(g) and Secure Communities programs and the enforcement actions they facilitate (e.g., immigration detainers, arrests, and removals) will reduce crime (e.g., OLC, 2002; Vaughan and Edwards, 2009). Given the substantial investments that have been made toward that end, it is important to assess whether these immigration law enforcement programs have, in fact, made America safer. The present research addresses this question.

Our study complements prior research on whether the adoption of 287(g) (Forrester and Nowrasteh, 2018), the Secure Communities program (Miles and Cox, 2014; Treyger et al., 2014), and anti-detainer policies (Ascherio, 2022; Gonzalez O’Brien, Collingwood, and El-Khatib, 2019; Hausman, 2020; Kubrin and Bartos, 2020; Martinez-Schuldt and Martinez, 2017; Wong, 2017) are associated with local crime rates in two significant ways. First, previous studies have generally examined local immigration enforcement policies in isolation when in practice the different policies (e.g., Section 287(g), Secure Communities, and sanctuary anti-detainer policies) have frequently been implemented side-by-side in one jurisdiction and their effects should be considered simultaneously. Secure Communities and Section 287(g) use data sharing and/or partnerships between ICE and local jurisdictions to ensure a faster and purportedly more efficient identification and removal of immigrants (Meissner, Kerwin, Chishti, and Bergeron, 2013). Sanctuary anti-detainer policies, in contrast, may counter these efforts by limiting cooperation with federal immigration officials (Martínez, Martínez-Schuldt, and Cantor, 2018). Many law review articles compare and address the legal and policy justifications of these policies (e.g., Lasch et al., 2018; Pham, 2018; Stumpf, 2015). It is, however, much harder to test the empirical relationship of these policies with crime, given that many jurisdictions are governed by multiple—and sometimes contradictory—federal and local policies. A city, for example, may pass ordinances to try to limit cooperation with federal detainers, but Secure Communities may still be operational in the city because local communities cannot opt out of the program (Dixon, 2020). A comprehensive examination of these policies requires a careful tracking of their implementation dates and factors related to implementation status for different jurisdictions, which is what the present study aims to do.

Second, previous studies have relied on crime data from the Uniform Crime Reporting (UCR) program to assess the relationship between local immigration policies and crime, which may provide an incomplete assessment due to known limitations of the UCR data. The UCR crime data are valuable for many purposes, but some features of the data raise significant questions about its utility for assessing the efficacy of local (e.g., city and county) immigration policies. Specifically, prior research has documented that the UCR omits a substantial amount of crime due to underreporting by citizens (Lynch and Addington, 2007) and non-reporting or partial reporting of crime data by law enforcement agencies (Maltz and Targonski, 2002; Targonski, 2011). There is evidence that these forms of missing data have changed over time and are correlated with community attributes relevant to variation in the activation of contemporary immigration policies (Baumer and Lauritsen, 2010; DeLang, Taheri, Hutchison, and Hawke, 2022; Martínez-Schuldt and Martínez, 2021; Xie, 2014; Xie and Baumer, 2019). Additionally, although Latinos have been the primary target of contemporary immigration policies (Martinez 2022) and, as delineated below, could be impacted by those policies uniquely, the UCR does not permit the analysis of crime exposure among distinct racial-ethnic groups. Thus, by restricting attention to UCR crime data, prior research has not been able to examine whether Latinos experience less, or more, crime after the activation of 287(g), Secure Communities, and anti-detainer policies.

The present study contributes to knowledge about the effects of 287(g), Secure Communities, and anti-detainer policies by examining their influence on victimization risk overall and among Latinos, non-Latino Whites, and non-Latino Blacks specifically. We go beyond previous research by considering all three policies in the same study and by using data on crime exposure from the U.S. National Crime Victimization Survey (NCVS). Although the NCVS has documented limitations as well (Lynch and Addington, 2007), it can provide a useful complement to the UCR for assessing the impact of local immigration policies because it includes crimes unknown to the police, is not affected by changes in local law enforcement practices, and permits analyses for different race-ethnic groups.

In the following sections we describe our research strategy and most important findings. Section 1 discusses background information and articulates the theoretical expectations for how county variation in the activation of immigration policies may affect exposure to crime. Section 2 reviews existing evidence and summarizes the key contributions of the present study. Section 3 describes the data and methods we use, and section 4 presents the results. The paper closes with section 5, where we summarize and discuss the most critical takeaways of the study.

1 |. BACKGROUND

1.1 |. Federal Immigration Enforcement Policies and Local Criminal Justice Priorities

The convergence of immigration and criminal law is known as “crimmigration” (Stumpf, 2006: 367), the roots of which extend back at least as far as the use of vice squad arrest records by the Immigration and Naturalization Service (INS) in the 1950s (Rosenbloom, 2015). The 1980s’ war on drugs, the more recent war on terror, and a myth of immigrant criminality provided much political momentum for the development of modern-day immigration enforcement measures like the Section 287(g) and Secure Communities programs, which are two key forms of policy initiatives specially designed to rely on the resources of local criminal law systems to expand the arrest and removal of immigrants in the interior United States (Chacón, 2009; Eagly, 2010).

Section 287(g).

Section 287(g) of the Immigration and Nationality Act, passed in 1996, allows ICE to enter agreements with local law enforcement agencies to allow these agencies to take on the roles of ICE agents and directly enforce federal immigration laws (Capps, Rosenblum, Rodríguez, and Chishti, 2011). It deputizes local law enforcement officials to perform immigration-related functions such as screening for immigration status, issuing detainers, holding individuals who violate immigration laws until the federal government takes custody, and allowing them to begin the process of removal. The first 287(g) agreement took effect in Florida in 2002, and for about a decade, local agencies could choose to implement a “jail enforcement” model (i.e., trained local officials check the immigration status of people who have been arrested and booked into local jails or correctional facilities), a “task force” model (i.e., local officials are authorized to question the immigration status of anyone they encounter in street-level policing), or a hybrid model that combines the two. ICE began to phase out the program in late 2012 as it came under intense scrutiny, when reviews of the program revealed evidence of insufficient oversight, racial profiling, and civil rights violations (Armenta, 2016; Coonan, 2013; Howerton, 2012). The revised 287(g) agreements under the Trump Administration kept only the jail enforcement model, while adding a new “warrant service officer (WSO)” model, permitting local officials to execute ICE administrative warrants within the local jails and correctional facilities. As of October 2022, ICE had 287(g) jail enforcement agreements with 64 law enforcement agencies in 19 states, and WSO agreements with 76 law enforcement agencies in 11 states (ICE, 2022a).

Secure Communities.

The Secure Communities program was launched in March 2008 (https://www.ice.gov/secure-communities). Under this program, when individuals are arrested and booked and their fingerprints are sent to the FBI for criminal background checks, their fingerprints are automatically transmitted to ICE. ICE then checks these fingerprints against their immigration databases to detect potential immigration violators and, when appropriate, issues detainers asking local agencies to hold the individuals for ICE to take custody. In issuing detainers, ICE officers making these decisions are instructed to prioritize the removal of individuals with criminal convictions who “pose the greatest threats to our communities” (ICE, 2008). The program was rolled out over a five-year period (Cox and Miles, 2013), and by January 2013 it was fully implemented in all 50 states, the District of Columbia, and five U.S. territories, totaling 3,181 county jurisdictions. It was suspended in November 2014 when a series of judicial decisions raised constitutionality questions about the use of immigration detainers (Manuel, 2015; Stumpf, 2015). In January 2017, it was re-implemented through an executive order by (then) President Trump, entitled Enhancing Public Safety in the Interior of the United States (Executive Office of the President, 2017), which remained in effect until it was revoked in January 2021 through an executive order issued by President Biden (Executive Office of the President, 2021).

Immigrant protective or “sanctuary” anti-detainer policies.

The functioning of the 287(g) and Secure Communities programs cannot be understood without reference to policies that are adopted by local jurisdictions to limit local participation in federal immigration enforcement. These policies can take many forms (Lasch et al., 2018; Martínez et al., 2018), but given our research interests, we focus on laws, ordinances, regulations, resolutions, or other practices that explicitly limit compliance with ICE detainer requests. Such policies may limit compliance for all detainer requests or permit compliance only if certain conditions are met, such as when the person named in the detainer has been convicted of a serious crime, the presence of an accompanying judicial warrant, or the presence of an agreement for ICE to reimburse the local agency for detention costs (CLINIC, 2014).¹ ICE detainers are a key immigration enforcement tool, as we described above. They are essential to the running of the 287(g) and Secure Communities programs, but when localities hold people in custody beyond their release date based on immigration detainers, they run a risk of litigation and financial liability for violation of the Fourth Amendment and due process rights, as many jurisdictions have discovered (ACLU, 2018).

The adoption of anti-detainer policies in local jurisdictions does not necessarily mean that the local jurisdictions are self-identified as “sanctuaries” for immigrants. That term is highly contested (McCormick, 2016), as many jurisdictions may have adopted the policies for practical reasons such as budgetary constraints and liability concerns, and not necessarily for reasons that indicate disagreement with federal policies or commitment to diversity and inclusion (Eagly, 2016; Herman, 2017; Martínez-Schuldt and Martínez, 2021). Regardless of the rationales, these policies allow local governments to allocate criminal justice resources based on local, rather than federal, priorities.

1.2 |. Debates about Potential Costs and Crime Reduction Benefits

Misplaced program priorities.

Secure Communities and 287(g) are focal points of contention among researchers and policymakers primarily because, despite stated priorities on serious felony offenders, both programs have deviated significantly from the stated goals by widening their intended net (Abrego, Coleman, Martinez, Menjivar, and Slack, 2017; Chacón, 2012; Ewing, Martínez, and Rumbaut, 2015). Based on an analysis of ICE records of 2.3 million deportations from 2008 to 2013, a report by Syracuse University’s Transactional Records Access Clearinghouse (TRAC, 2014a) found that 52% of deportees in the data had no criminal conviction at all; and that of those with criminal convictions, the most serious conviction was an immigration or traffic violation in 41% of the cases. The expansion of the Secure Communities program during that period (2008–2013) coincided with a decline in the percentage of deportees with more serious criminal convictions, not an increase as one would expect from the program based on its stated purpose and intent (TRAC, 2014a). Meanwhile, a similar pattern of findings was observed among studies of the Section 287(g) program that showed that police officers in active 287(g) jurisdictions often use the authority to focus on minor infractions, such as misdemeanor traffic offenders who “drive while Brown” (or Black) (Coleman, 2012; Coleman and Stuesse, 2016; Lacayo, 2010). As elaborated below, these practices raised questions about the capacity of Secure Communities and 287(g) to effectively reduce crime, and many scholars have expressed doubts on this matter (e.g., Kubrin, 2014; Lyons, Vélez, and Santoro, 2013; Martinez and Iwama, 2014; Provine and Varsanyi, 2012; Rosenfeld, 2014; Walker and Leitner, 2011; Valdez, Coleman, and Akbar, 2017). Researchers also have suggested that crimes may actually go up with greater local involvement in immigration enforcement under the 287(g) and Secure Communities programs, by the mechanisms such as the disruption of local social and economic life, the endangerment of local crime-fighting resources, the damages to community cooperation with law enforcement, and the increased vulnerability of immigrants and racial-ethnic minorities as crime victims (Forrester and Nowrasteh, 2018; Nguyen and Gill, 2016), as described below.

Social and economic consequences.

The 287(g) and Secure Communities programs, even when they are administered in jails, open the door for broader law enforcement because any officer, not just those working in jails, may arrest someone suspected of immigration violations knowing that their immigration status will be checked when they are in jails (Coleman, 2012; Gardner and Kohli, 2009). Not surprisingly, such practices have been shown to have widespread effects on social and economic activities in the communities involved. A review by Castañeda et al. (2015), for example, documented the adverse effects of immigration enforcement activities such as raids, detention, deportation, and family separation on health and well-being of both immigrants and U.S. citizens (also see Brabeck, Lykes, and Hunter, 2014; Dreby, 2012). Moreover, heightened immigration enforcement incurs high costs to communities, as local governments must bear the financial burdens of such activities, which are estimated to be millions of dollars each year, in operational costs alone, for many U.S. counties (Nguyen and Gill, 2010). Beyond these costs, there are potential indirect costs that are associated with the criminalization of immigrants, such as the risk of litigation and liability that local communities may face (Stumpf, 2015), loss of business and tax revenue (Lacayo, 2012), a lower supply of immigrant workers in certain jurisdictions (Parrado, 2012), and higher costs of services and goods due to less available inexpensive labor (Kasarda and Johnson, 2006).

Community-police relationships.

Local authorities, seeing the high costs of immigration enforcement, may exercise their own initiative by declining to use local law enforcement agencies in pursuit of such efforts (Martínez et al., 2018). Some are not dissuaded by the costs and are ready to use local resources in the enforcement of immigration laws. In places in Arizona, Texas, Florida, Alabama, North Carolina, and other states in southeastern United States, for example, where the restrictive federal policies have been enforced with vigor, researchers have noted that the enforcement efforts create an inhospitable environment that may reduce police legitimacy (Arriaga, 2017; Gardner and Kohli, 2009; Menjívar, 2016; Parrado, 2012). For instance, research findings from fieldwork carried out in Latino and Black immigrant communities have found deteriorating relationships between residents and the police, and reluctance of residents to report crime and provide information as witnesses (e.g., Hagan, Castro, and Rodriguez, 2010; Menjívar and Bejarano, 2004; Rengifo and Fratello, 2015; Theodore and Habans, 2016). These fieldwork data are consistent with national and sub-national surveys of Latinos (regardless of immigration status or national origin) who reported substantial concerns about police involvement in immigration enforcement (e.g., Pew Hispanic Center, 2007; Theodore, 2013). The data from the NCVS similarly showed a substantial decline in victims’ willingness to call the police to report crime for those living in immigrant-concentrated neighborhoods that are located inside new immigrant destinations, where the tendency for local jurisdictions to adopt restrictive immigration enforcement policies is particularly strong (Xie and Baumer, 2019).

Expected impacts on crime.

The literature has emphasized highly divergent streams of thought on the anticipated impact of immigration policies on crime. Proponents of 287(g) and the Secure Communities program suggested that the adoption of these policies would yield public safety benefits through a combination of deterrence and incapacitation that would reduce crime (e.g., OLC, 2002; U.S. Congressional House Committee on Homeland Security, 2009; Vaughan and Edwards, 2009). In contrast, many scholars and policymakers warned that these programs could fuel distrust of the police, contribute to a culture of legal cynicism within immigrant communities (e.g., Donato and Rodriguez, 2014; GAO, 2009; Homeland Security Advisory Council, 2011; Kubrin, 2014; Lyons et al., 2013; Martinez and Iwama, 2014; Muchow and Amuedo-Dorantes, 2020), and reduce crime reporting and cooperation with the police (Aguilasocho, Rodwin, and Ashar, 2012; Decker, Lewis, Provine, and Varsanyi, 2009; Menjívar and Abrego, 2012; Theodore and Habans, 2016; Wong et al., 2019). Additionally, research indicates that the activation of 287(g) and Secure Communities, and the enforcement actions they facilitate, may lead to significant family disruption and economic hardship in immigrant communities (Artiga and Lyons, 2018; Capps et al., 2015). These conditions may not only mitigate any expected crime reduction but also could lead to crime increases (Kirk et al., 2012; Kubrin, 2014; Martinez and Iwama, 2014; Zatz and Smith, 2012).

A similar divergence of views has emerged for the anticipated effects of anti-detainer policies on crime (Gamboa, 2015). Proponents of anti-detainer policies suggest that they may enhance public safety by improving community trust in the police (International Association of Chiefs of Police, 2005), whereas opponents suggest that they may result in protection for criminal activities, which could yield higher levels of crime (see Gonzalez O’Brien, Collingwood, and El-Khatib, 2019, for a review of these contrasting views).

Many discussions about the potential impacts of 287(g), Secure Communities, and anti-detainer policies on public safety, and all the empirical studies to date (reviewed below), have focused on overall crime without explicit references to race-ethnicity. While this is a useful starting point, these policies have been designed and implemented largely to disproportionately affect Latinos (Martinez, 2022). Latinos, for example, represent a large majority of persons arrested, detained, or deported through 287(g) and Secure Communities.² This reality, coupled with the strong intra-group dynamics of crime in the US, implies that the impact of these policies is likely to differ significantly by race/ethnicity. Specifically, as Latinos are the primary focus of immigration enforcement (TRAC, 2014b), they could be particularly affected by the policies, although for reasons noted above the direction of the anticipated effects are theoretically ambiguous. In contrast, given that non-Latino Whites and Blacks are not the main subject of immigration enforcement efforts, these policies should have weaker effects on their exposure to crime. We next summarize the existing empirical evidence on these divergent predictions about the impact of local immigration policies on crime.

2 |. EXISTING EVIDENCE AND THE CONTRIBUTION OF THE PRESENT STUDY

2.1 |. Previous Research Strategies

A small body of research has capitalized on variation in the timing of the rollout of the Secure Communities program across U.S. counties to assess whether the activation of this policy influenced crime. Miles and Cox (2014) used monthly, county-level UCR data from 2004 to 2012 to do so, and their estimation of two-way fixed effects models indicated that the activation of the Secure Communities program showed no significant association with UCR county crime rates. Using similar methods, Treyger et al. (2014) reached a parallel conclusion in their study, which showed that Secure Communities had no significant influence on monthly UCR rates of offenses and arrests for 335 cities from 2008 to 2011. More recently, Kang and Song (2022) reported similar null effects of Secure Communities activation. Kang and Song (2022) presented additional evidence suggesting a deterrent effect of Secure Communities when it was also activated simultaneously in nearby areas, though it is unclear whether the pattern persists while controlling for the spatio-temporal dependence of crime or the spatial effects of other predictors.

To our knowledge, one previous study has considered the impacts of 287(g) on crime. Forrester and Nowrasteh (2018) considered the effect of the 287(g) program in North Carolina using annual UCR crime rates in 100 counties from 2003 to 2013. Using a staggered difference-in-differences research design, Forrester and Nowrasteh (2018) found that the 287(g) program had no significant effect on UCR crime rates. However, they observed that the program increased the number of assaults against police officers, which is consistent with findings from other research about the negative consequences of the 287(g) program for community-police relations (Nguyen and Gill, 2010).

Beyond the studies of the 287(g) and Secure Communities programs, research on the impact of sanctuary policies on crime has grown. Using cross-sectional data, Lyons et al. (2013) found that neighborhoods in cities with at least one law or formal resolution limiting local enforcement of immigration laws had lower rates of homicide and robbery than neighborhoods in “non-sanctuary cities,” net of the effects of other neighborhood and city characteristics. A county-level study also reported evidence that UCR crime rates are lower in counties with anti-detainer policies compared to counties without them, by matching the counties’ population characteristics (Wong, 2017). The results for sanctuary policies are less consistent in longitudinal studies. Martinez-Schuldt and Martinez (2017) analyzed data from 107 cities observed in 1990, 2000, and between 2005 and 2010, finding that the adoption of sanctuary policies is associated with a reduction in robberies but not homicide in models that included city fixed effects. In contrast, Gonzalez O’Brien et al. (2019) found that sanctuary city policies showed no demonstrable effect on UCR crime rates in 20 states and the District of Columbia between 2000 and 2014. Similar null findings were reported by Hausman (2020) in a county-level study of UCR crime data from 2010–2015, and by Kubrin and Barto’s (2020) study based on their analysis of California’s sanctuary state policy (Senate Bill 54) and its impact on statewide UCR crime rates. Finally, more recent findings suggest that the impact of sanctuary policies on crime may be contingent on offense type and the time period studied. Ascherio (2022) reported results that showed no impact of sanctuary policies on crime through 2012, but significantly greater reductions in UCR robbery and burglary rates (but not other crimes) in sanctuary counties from 2013 to 2016. Similarly, Manning and Burkhardt’s (2022) fixed effects analysis of UCR crime rates from 2010–2016 found larger reductions in county-level rates of aggravated assault (but not other crimes) after 2014 in counties that adopted sanctuary policies.

In summary, the weight of the existing empirical evidence suggests that the activation of 287(g), Secure Communities, and anti-detainer policies has not shown clear evidence of a significant association with public safety. Yet, as referenced in the introduction, two features of prior research motivate additional investigation using alternative methodological approaches.

First, we have suggested that those closely related immigration enforcement policies (Secure Communities, Section 287(g), and anti-detainer policies) should be evaluated together, given that they have been frequently implemented side-by-side in the same jurisdictions, sometimes with contradictory purposes. Secure Communities and Section 287(g) both aim to use local law enforcement resources to achieve a purportedly faster and more efficient immigration control system (Meissner, Kerwin, Chishti, and Bergeron, 2013). Sanctuary anti-detainer policies, in contrast, may counter these efforts by limiting cooperation with federal immigration officials (Martínez, Martínez-Schuldt, and Cantor, 2018). From a legal and constitutional perspective, many scholars have compared the legitimacy and inner workings of these interconnected policies on theoretical grounds (e.g., Lasch et al., 2018; Pham, 2018; Stumpf, 2015). Empirically, in order to test the relationships of these policies to crime, we extend the previous studies by tracking the different policies’ implementation dates and factors related to their implementation in jurisdictions across the U.S. in order to gain a more comprehensive view of how different approaches to immigration enforcement may work.

Second, we also have noted the benefits of expanding the UCR or police-based research to analyses using the NCVS data. While the UCR crime data are valuable for many purposes, several features of the data have limited their utility for assessing the efficacy of immigration enforcement policies in reducing crime. Perhaps most notably, as we reasoned above, 287(g), Secure Communities, and anti-detainer policies may have varying effects on crimes for selected racial/ethnic groups, with anticipated impacts most prominent for Latinos. Unfortunately, a limitation of the UCR data is that they lack the capacity to analyze crime disaggregated by race/ethnicity, and as a consequence, none of the studies reviewed has considered impacts on exposure to crime for Latinos and their non-Latino peers. What is more, researchers have noted other limitations of the UCR data that may hinder the evaluation of local immigration enforcement policies, including missing data problems in the UCR and other police-related factors for shaping the UCR data. Many law enforcement agencies do not submit data to the UCR program, or they may submit data for some but not all months within a given year (Maltz and Targonski, 2002; Targonski, 2011). These forms of missing data in the UCR pose challenges for estimating annual and monthly crime rates for cities and counties and may yield community samples that are not representative of the underlying populations of interest because some states and many local areas are missing from the analysis. Furthermore, recent research has reported evidence that missing data in the UCR is significantly correlated with a wide range of attributes of local communities, including racial composition (DeLang et al., 2022), which may affect the implementation and evaluation of immigration enforcement policies. Relevant to our study, for example, there is evidence that crime experienced in immigrant communities is less likely to be reported to the police (Xie and Baumer, 2019), that some immigration policies (i.e., anti-detainer policies) also affect whether the public contacts the police to report crime (Martínez-Schuldt and Martínez, 2021), and that areas with high levels of immigration enforcement may be over-policed compared with other places (Kubrin, 2014).

In summary, valuable insights have been gained from the analyses of UCR data, but we suggest that it is desirable to consider alternative data sources on crime to examine the impact of contemporary immigration policies. The NCVS offers one such data source that offers an important complement to the UCR.

2.2 |. The Present Study

Our investigation integrates data from the NCVS with multiple other sources to advance knowledge on the impact of immigration enforcement policies on violent victimization risk among U.S. residents. Unlike the macro-level studies reviewed above, where the unit of analysis is a geographic area or a political administrative unit, the data used in our study are multilevel, with repeated measures of individuals nested within census tracts and counties. This approach enables us to examine changes in individual victimization risks over time using a within-person design. If the public safety rationales frequently articulated as the justification for 287(g), Secure Communities, and anti-detainer policies are valid, then the activation of such programs should lead to a reduced risk of violent victimization. However, if critics are right about the adverse impacts of these policies, we may observe higher levels of victimization following their activation.

3 |. DATA AND METHODS

3.1 |. Data

The study used data compiled from multiple sources. The main data were drawn from the Bureau of Justice Statistics’ (BJS) National Crime Victimization Survey (NCVS) from 2005 through 2014, which were linked with restricted-use census tract and county codes in the NCVS Address File that the Census Bureau maintains. The NCVS is the largest victimization survey in the United States, sponsored by the BJS with the intention to create a more accurate measure of crime by using self-report data from victims to yield information on crime incidents, whether or not they have been reported to the police (Lynch and Addington, 2007). Each year, the Census Bureau administers the NCVS to a nationally representative sample of approximately 90,000 households comprising about 160,000 individuals aged 12 years or older, with a survey response rate above 85% for the years under study (Planty, 2014). The survey uses a rotating panel design in which sample households are interviewed at intervals of six months for up to seven interviews before exiting the survey. The questionnaire asks about respondents’ victimization experiences during the six-month recall period, as well as information on their personal and household characteristics.

The NCVS is useful for our purposes because it includes victimizations not reported to the police and permits analysis of victimization by race-ethnicity. Like the UCR, however, it has limitations that should be considered. By design, the NCVS excludes lethal violence, crimes against businesses, and crime experiences from persons unattached to residential housing units and those under the age of 12. Additionally, research suggests that rates of survey response in the NCVS differ by age, race-ethnicity, and selected other individual and household attributes (Bureau of Justice Statistics, 2014; Cohen and Lynch, 2007; Lohr, 2019). The NCVS contains weights that effectively adjust for these sources of nonresponse (Zhang, Woodburn, and Scheuren, 2009). However, panel attrition in the NCVS may be correlated with changes in county immigration policies, which may not be accounted for by the sample weights, and therefore we examine this issue in our analysis, implementing appropriate adjustments to minimize potential bias. Finally, the activation of immigration policies in a community may affect not only the likelihood that people report crime to the police (Martínez-Schuldt and Martínez, 2021), but also whether NCVS respondents in such areas acknowledge victimizations to survey administrators.

To facilitate the linkage of data on local immigration policy implementation to the NCVS records, we accessed restricted-use geographic codes for the respondents’ residential counties and census tracts and conducted the analyses within approved Federal Statistical Research Data Centers (FSRDCs). We focused on the 2005–2014 period to capture observations before and after the activation of the Secure Communities program in U.S. counties from 2008 to 2013 (Cox and Miles, 2013) and to incorporate the first major wave of 287(g) agreements implemented (Forrester and Nowrasteh, 2018).³ Focusing on this study period parallels existing research that has relied on police-based crime data.

We used several publicly available data sources to compile information on immigration enforcement policies in U.S. counties about the activation of Secure Communities, 287(g) agreements, and anti-detainer measures adopted for the years under study. Specifically, Secure Communities data were obtained from the ICE’s Freedom of Information Act (FOIA) library. This library includes Congressional Status Reports and Secure Communities Interoperability Statistics, which recorded dates when Secure Communities programs were activated (ICE, 2021).⁴ Data on the presence of 287(g) agreements and the dates they were active were also obtained from the ICE FOIA library and validated through a comparison with other sources (Capps et al., 2011; Gelatt, Bernstein, and Koball, 2017). Data on anti-detainer policies and dates of enactment were gathered and cross-checked from four different sources: ICE’s “Weekly Declined Detainer Outcome Report” (ICE, 2017), the House Appropriations Committee’s reports on the Department of Homeland Security Appropriations Bill for fiscal years 2007 through 2018 (U.S. House Committee on Appropriations, 2006–2017), the Catholic Immigration Legal Network’s (CLINIC, 2014) list of anti-detainer policies, and the interactive map data from the Center for Immigration Studies (Griffith and Vaughan, 2017).

Our assessment of the impact of county-level immigration policies on victimization risk also accounted for other community attributes. We used data from the American Community Survey (ACS), Decennial Census, and Bureau of Labor Statistics (BLS) to measure socioeconomic and demographic characteristics of census tracts and counties that prior research suggests may influence law enforcement functioning and crime rates (King, Messner, and Baller, 2009). Additionally, we used data from the Census of Law Enforcement Agencies and the Law Enforcement Agency Roster supplied by the BJS and the Census of Governments to compile data on the varying levels of law enforcement resources across U.S. counties (Pierson, Hand, and Thompson, 2015; Xie and Baumer, 2019).

The integration of these data sources yielded a nationally representative sample of 354,000 U.S. residents aged 12 and older with a total of 1,006,000 interviews,⁵ with time varying information on county immigration enforcement policies and other information measured at the individual, census tract, and county levels, as detailed next.⁶

3.2 |. Dependent Variable

The dependent variable for our analyses was a dichotomous measure of whether a respondent experienced a violent victimization (including rape, sexual assault, robbery, and assault) during the six months preceding the interview. Violence is of keen public interest (Sharkey, 2018), and it is a focus of much of the public dialogue about immigration policy and crime.⁷ We used a binary outcome variable (1=yes, 0=no) rather than victimization counts because few respondents reported multiple victimizations within the six-month reference period.

3.3 |. County Immigration Enforcement Policies

Secure Communities Program.

We included a time-varying dummy variable to identify whether Secure Communities was active in the respondent’s county (Secure Communities Activated) during the six-month reference period (1=yes, 0=no). In some instances, Secure Communities was active for only part of a respondent’s reference period. For these cases (less than 5% of cases), we tested the robustness of the results by coding the variable as the proportion of the period that the program was active. We found that the results were not sensitive to the coding scheme or to the decision of whether to exclude these cases from the analyses.

Section 287(g) Jail Enforcement or Task Force Agreements.

We used two variables to represent whether a 287(g) agreement was in effect for the respondent’s county during the six-month reference period in either jail enforcement or task force settings. Counties that had a hybrid model, which included both the jail enforcement and task force models, were assigned a value of 1 for both variables. For cases where the 287(g) agreements were active for part of the reference period (less than 1% of the sample for both types of agreements), we also tested the robustness of the analyses by coding the variables to indicate the proportion of the reference period that the agreements existed. The coding of these cases and whether we included or excluded these cases had no impact on the findings reported.

Anti-detainer policies.

Our last policy variable indicates whether the respondent’s county had implemented anti-detainer policies limiting cooperation with ICE during the reference period. Such policies may prohibit local law enforcement agencies from honoring ICE detainer requests on all occasions, or they may limit compliance to cases only when ICE has a judicial warrant or court order to support the detention, when ICE has agreed to reimburse the local agencies for the detention costs, or when the subject of the request has been convicted of a certain felony or serious crime (CLINIC, 2014). Preliminary analyses indicated no discernable differences in effects of these forms of anti-detainer policies on victimization, so we combined them into a single indicator. When an anti-detainer policy was active for only part of the reference period (1.4% of the sample), we coded the variable with the proportion of the reference period for which the policy existed. Substantively identical findings were observed with this modified coding, and whether we included or omitted these cases for the analyses.

3.4 |. Control Variables

Individual Risk Factors for Victimization.

Prior studies have suggested that a person’s age, gender, race/ethnicity (Latino, non-Latino White, Black, and other race), marital status, education, employment status, income, homeownership (owned or rented), and residential tenure are key indicators of social vulnerability, lifestyles, and opportunity structures that influence victimization (e.g., Dugan and Apel, 2003; Fisher and Cullen, 2000; Lauritsen and Rezey, 2018). Drawing from the NCVS, we included measures of each of these individual-level attributes. We also included survey administration variables that may be related to victimization risk. In the NCVS, newcomers with residence durations shorter than six months had less time at risk for victimization than did other residents. We therefore included survey reference time to control for these exposure differences. Additionally, repeated interviews may lead to reduced reporting of victimization in the NCVS (Hart, Rennison, and Gibson, 2005; Lynch, Berbaum, and Planty, 2002), so we included time in sample to allow for respondent fatigue and testing effects. We also controlled for year of interview (10 categories for 2005–2014, with year 2005 used as the reference category) to adjust for possible period effects. The detailed coding of these individual-level control variables is presented in Appendix A.

Neighborhood Risk Factors for Victimization.

At the neighborhood level (measured by census tracts), we controlled for neighborhood social and economic conditions including population density, concentrated disadvantage, racial/ethnic heterogeneity, neighborhood age profile, family disruption, and residential instability, which are important contextual factors for victimization according to social disorganization and routine activity theories (Sampson and Lauritsen, 1994).⁸ Specifically, neighborhood population density (logged population per square mile) and a binary indicator of the neighborhood’s location inside a central city were used to account for the higher concentration of crime in urban and more densely populated areas (Glaeser and Sacerdote, 1999). Victimization rates also have been shown to be higher in more disadvantaged neighborhoods (Sampson, Raudenbush, and Earls, 1997), which we accounted for by including a composite disadvantage index combining poverty rate, unemployment rate, percent female-headed households with children, median household income adjusted for inflation (sign reversed), percent households with public assistance income, and percent population 25 years and over without a high school diploma (coefficient alpha=.89). Neighborhood differences in racial/ethnic composition were measured by the percentages of racial/ethnic minorities (Black, Latino, Asian and Pacific Islander, and other group) and by the entropy index of the level of multigroup diversity in neighborhoods (Hipp, 2011). We also measured the percentage of foreign-born residents, which has been found in previous studies to influence neighborhood social cohesion and economic change and in turn influence victimization (Martinez, Stowell, and Lee, 2010; Ousey and Kubrin, 2018; Xie and Baumer, 2018). The percentage of the population aged 18–34 was included to control for potential differences across neighborhoods in youthful activity and risk for criminal violence (Messner, Lu, Zhang, and Liu, 2007). Neighborhood family disruption and residential instability were measured by three indicators: percent divorced/separated, percent households that moved into their unit less than 10 years ago, and percent vacant housing (Logan, Xu, and Stults, 2014). By disrupting family relationships and social cohesion among residents, these variables may contribute to higher levels of victimization (Sampson, 2017; Xie and McDowall, 2008).

County-Level Control Variables.

Our models also included measures of a county’s labor market condition (percent unemployment), immigration history (traditional or new immigrant destinations), policing resources (police force size and expenditures), adjacency to the border (with Mexico or Canada), and regional location (i.e., South, Midwest, West, and Northeast).⁹ Ethnic competition and scapegoating arguments suggest that localities experiencing greater unemployment may develop more anti-immigrant sentiment and exclusionary policies (Cochrane and Nevitte, 2014). This relationship, along with the theoretical expectation that unemployment rates may affect crime (Andresen, 2015; Bell, Bindler, and Machin, 2018), suggests the need for controlling county unemployment rates in our analyses. Regional dummies and type of immigrant destinations are important control variables as well, for they are related not only to victimization rates (Harris and Feldmeyer, 2013; Ramey, 2013; Xie and Baumer, 2018), but also to the decision of a locality to adopt a certain immigration policy, as localities in the South and new immigrant destinations have been shown to be more likely than other localities to implement stringent immigration enforcement policies (Walker and Leitner, 2011; Winders, 2007). We used the typology proposed by Suro and Singer (2002) to classify counties by immigration history: Traditional immigrant counties are those with long-established immigrant populations that exceeded the national average of 7.9% in 1990 (see map in Appendix B; this category includes Los Angeles, Chicago, New York City, Miami, and other well-established immigrant gateway counties). Non-traditional counties include two categories: new immigrant counties experienced relatively large growth in immigrant population that exceeded the national average growth rate of 102% from 1990 to 2010, whereas small immigrant counties did not. As Appendix B shows, new immigrant areas were disproportionately clustered in the Southeast and West, whereas small immigrant areas were disproportionately clustered in the North and parts of the South. Because of the differences in the rate of immigrant population growth and differences in resource availability, traditional, new, and small immigrant counties may face different opportunities and challenges in housing immigrants (Waters and Jimenez, 2005). These differences may translate into differences in victimization patterns as well as different approaches to immigration law enforcement (e.g., Pierotte et al., 2018; Shihadeh and Barranco, 2013). We therefore included these variables as controls in our analyses.

The available police resources (measured by the number of sworn police officers per 1,000 population and per capita police expenditures) have been theorized in prior studies to affect the amount of police efforts devoted to investigating suspicious circumstances and activities that may help reduce crime, although the literature is still mixed about the mechanisms (e.g., Chalfin and McCrary, 2018; Levitt, 2002; Loftin and McDowall, 1982; Mello, 2019; Owens, 2019). Finally, two binary variables were included to control for differences between counties along the Northern and Southern borders and counties in the interior United States in both violence rates (Orrenius and Coronado, 2005) and in resources and law enforcement activities devoted to enforcing national borders and interior immigration laws (Amuedo-Dorantes and Pozo, 2014).

3.5 |. Analytical Strategy

We employed multiple analysis strategies to evaluate the impact of immigration enforcement policies on victimization. To maximize the use of the available data, and to test the extent to which both between-person and within-person variations in county immigration enforcement policies affect victimization risks, we first present results from a longitudinal “hybrid” (i.e., the “between-within”) model (Allison, 2009; Neuhaus and Kalbfleisch, 1998; Raudenbush and Bryk, 2002) that used repeated interviews with the same respondents and group-mean centering to partition the policy effects into two components: between-person difference and within-person change. This model takes the following general form:

$$\begin{gathered} log\left(\frac{\mathrm{Pr}\left(Y_{wijk}=1\right)}{1-\mathrm{Pr}\left(Y_{wijk}=1\right)}\right)={\beta }_{0000}+{\beta }_{11}{\overline{ImmigrationEnforcementPolicies}}_i \\ +{\beta }_{12}\mathrm{\Delta }ImmigrationEnforcementPolicie{s}_{wi} \\ +{\beta }_{21}{\overline{IndividualAttributes}}_i+{\beta }_{22}\mathrm{\Delta }IndividualAttribute{s}_{wi} \\ +{\beta }_{31}{\overline{NeighborhoodAttributes}}_i+{\beta }_{32}\mathrm{\Delta }NeighborhoodAttribute{s}_{wi} \\ +{\beta }_{41}{\overline{OtherCountyAttributes}}_i+{\beta }_{42}\mathrm{\Delta }OtherCountyAttribute{s}_{wi} \\ +{\beta }_{5}Yea{r}_D+r_{0ijk}+U_{00jk}+V_{000k} \end{gathered}$$ (eq.1)

As the notation reveals, equation 1 includes many time-varying individual, neighborhood, and county explanatory variables, along with a few time-invariant attributes (e.g., sex, race-ethnicity, region, and residence in a county adjacent to the U.S. borders) and a person-specific error term $\left(r_{0ijk}\right)$ that is assumed to be normally distributed with mean zero and variance ${\sigma }_r^2$ to allow for persistent heterogeneity in the average level of victimization across persons in the sample. The key parameters of the model are ${\beta }_{11}$ (estimates of between-person associations between immigration policies and victimization) and ${\beta }_{12}$ (estimates of within-person associations between change in immigration policies with change in victimization risk over time). The estimation of both parameters makes the hybrid (between- and within-person) approach attractive for our research purposes, as it provides estimates that parallel the outputs of previous cross-sectional and longitudinal research on immigration policy and crime as measured from police data. Nonetheless, our conclusions about the impact of county immigration policies on victimization risk focus on the within-person estimates, which account for unobserved time-invariant predictors and provide a more rigorous assessment of causal impacts (Firebaugh, Warner, and Massoglia, 2013).¹⁰

To ensure that the within-person results obtained from the hybrid design are robust to alternative estimation approaches, we also present results from fixed-effects difference-in-differences (DID) logistic regression models that base estimates only on the within-person variance (Wooldridge, 2013). We assessed the credibility of the parallel trends assumption of this design in supplementary analyses using lead-lag indicators. These supplementary analyses integrated leads and lags of the policy activation variables to the fixed effects specification reported below, yielding a dynamic DID model, also referenced in the statistical literature as a DID event study (see, e.g., Clarke and Tapia-Schythe, 2021; Goodman-Bacon, 2021). For all policies examined, we found that the lead coefficients were statistically indistinguishable from zero, which supports the parallel trends assumption. The policy lag coefficients were also in line with the results reported below. We further assessed whether the results were sensitive to the exclusion of time points that were relatively far away from the time of policy activation and the exclusion of very-early or very-late activation counties (Clarke and Tapia-Schythe, 2021).¹¹ Specifically, we re-estimated the models using only respondents that have data for the middle periods (e.g., from leads 4 to lags 4) and samples that exclude those from early-treated counties and late-treated counties.¹² We found that the conclusions reported below did not change with these different model specifications.¹³

4 |. RESULTS

4.1 |. Analyses of the Full Sample of All Race/Ethnicities

Table 1 shows descriptive statistics for the sample of persons aged 12 and older who were surveyed about experiences with violence from 2005 to 2014. On average, by pooling the data of all the years together, about 7 in 1,000 persons reported experiencing violent victimizations in their own counties within the preceding 6-months. About 43% of respondents were from counties in which the Secure Communities program was operational at the time. The other policies were active for a much smaller proportion of the sample, but since the sample is large the data enable us to make meaningful comparisons among those who did or did not experience the activation of these policies in their communities. The 287(g) “jail enforcement” and “task force” programs were affecting about 11% and 3% of respondents, respectively. Additionally, about 3% of respondents had anti-detainer policies in their counties.

Table 1.

Descriptive statistics of study variables, 2005–2014

Characteristics	Mean	(SD)
Dependent variable
Violent victimization	.007	(.074)
County policy variables
Secure Communities activated	.426	(.433)
287(g) jail enforcement agreement	.106	(.273)
287(g) task force agreement	.026	(.140)
Anti-detainer policy activated	.031	(.147)
Control variables
Individual-level variables
Age	45.330	(16.580)
Male	.488	(.446)
Latino	.135	(.304)
Black	.110	(.280)
Other nonwhite	.055	(.203)
Divorced	.106	(.274)
Separated	.021	(.127)
Never married	.297	(.407)
Education	13.060	(2.591)
Employed	.626	(.432)
Household income	10.790	(3.245)
Homeowner	.692	(.412)
Length of residence	11.550	(11.020)
Neighborhood-level variables
Population density	7.083	(1.783)
Central city neighborhood	.306	(.411)
SES disadvantage	−.010	(.661)
% Black	12.090	(17.270)
% Latino	14.270	(17.950)
% Asian and Pacific Islander	4.717	(7.040)
% other race/ethnicity	1.374	(2.760)
Race entropy	.672	(.300)
% foreign born	11.500	(11.570)
% ages 18–34	23.090	(8.491)
% divorced/separated	12.870	(4.077)
% moved into units <10 years	59.190	(11.850)
% vacant housing	9.736	(6.958)
County context
% unemployed	7.269	(2.472)
Southern border with Mexico	.037	(.168)
Northern border with Canada	.031	(.154)
Traditional immigrant county	.329	(.419)
New immigrant county	.372	(.431)
Police force size	2.009	(.686)
Police expenditures	152	(75.880)
South	.362	(.429)
Midwest	.245	(.384)
West	.221	(.370)
Survey administration variables
Survey reference time	5.766	(.822)
Time in sample	2.748	(1.517)
Year of interview
2006	.080	(.241)
2007	.077	(.238)
2008	.108	(.276)
2009	.108	(.277)
2010	.109	(.277)
2011	.109	(.278)
2012	.113	(.282)
2013	.113	(.283)
2014	.114	(.283)
Number of persons	354,000
Number of person-interviews	1,006,000

Note. SD = standard deviation. The summary statistics were weighted to represent the U.S. population.

Table 2 presents results from the longitudinal “hybrid” logistic regression model. The between-person results confirm previous studies that have identified person-level and neighborhood-related factors as major sources of variation in violent victimizations (Lauritsen and Rezey, 2018). A person’s victimization risk, for example, is found to be significantly higher among males, younger residents, divorced, separated or other unmarried people, lower-income residents, and residents who reside in densely populated, central city, and more socioeconomically disadvantaged areas (for similar findings, see Lauritsen and White, 2001; Schreck and Fisher, 2004; Wright and Benson, 2011). The results also show a regional difference, with counties in the West showing a higher risk of victimization than those in the Northeast. Survey administration variables are important as well, reaffirming findings in other studies about the effects of “survey reference time” and “time in sample” on victimization risk and, as expected in light of national crime trends observed during the period, the “year of interview” slopes generally show a decline in victimization risk in later years of the survey (2009–2014) compared with 2005.

Table 2.

Estimated Effects of County Immigration Enforcement Policies on Violent Victimization, 2005–2014

	Longitudinal hybrid model
	Between-person difference			Within-person change
Characteristics	b		(SE)	b		(SE)
County policy variables
Secure Communities activated	.076		(.076)	.151	*	(.069)
287(g) jail enforcement agreement	−.072		(.078)	.339		(.256)
287(g) task force agreement	.226		(.131)	.305		(.217)
Anti-detainer policy activated	.017		(.129)	.049		(.199)
Control variables
Individual-level variables
Age	−.025	***	(.001)	.032		(.033)
Male	.173	***	(.028)	--		--
Latino	−.326	***	(.059)	--		--
Black	−.291	***	(.053)	--		--
Other nonwhite	−.137		(.079)	--		--
Divorced	.850	***	(.040)	.068		(.241)
Separated	1.129	***	(.063)	1.061	***	(.296)
Never married	.338	***	(.038)	−.179		(.197)
Education	−.004		(.006)	.014		(.035)
Employed	−.070	*	(.033)	.096		(.068)
Household income	−.060	***	(.005)	.009		(.008)
Homeowner	−.126	***	(.035)	.009		(.203)
Length of residence	−.011	***	(.002)	.008		(.008)
Neighborhood-level variables
Population density	.034	*	(.014)	.497		(.580)
Central city neighborhood	.087	*	(.040)	--		--
SES disadvantage	.160	***	(.036)	−.247		(.231)
% Black	.001		(.001)	−.015		(.055)
% Latino	.001		(.002)	−.028		(.048)
% Asian and Pacific Islander	−.014	**	(.005)	.138		(.128)
% other race/ethnicity	.008	*	(.004)	−.061		(.248)
Race entropy	.152	*	(.066)	−.209		(.666)
% foreign born	−.004		(.003)	−.024		(.018)
% ages 18–34	−.006		(.004)	.021		(.015)
% divorced/separated	.011	**	(.004)	−.014		(.019)
% moved into units <10 years	.003		(.002)	−.008		(.011)
% vacant housing	.006	*	(.002)	.015		(.013)
County context
% unemployed	−.010		(.011)	.003		(.018)
Southern border with Mexico	−.250		(.135)	--		--
Northern border with Canada	.058		(.109)	--		--
Traditional immigrant county	−.048		(.068)	--		--
New immigrant county	−.024		(.056)	--		--
Police force size	.013		(.039)	.687		(.422)
Police expenditures	.000		(.000)	−.002		(.003)
South	−.159		(.082)	--		--
Midwest	.012		(.069)	--		--
West	.224	**	(.078)	--		--
Survey administration variables
Survey reference time	.209	***	(.015)	--		--
Time in sample	−.229	***	(.015)	--		--
Year of interview
2006	−.071		(.074)	--		--
2007	−.093		(.069)	--		--
2008	−.081		(.067)	--		--
2009	−.205	**	(.074)	--		--
2010	−.224	**	(.086)	--		--
2011	−.192	*	(.092)	--		--
2012	−.179		(.098)	--		--
2013	−.343	***	(.102)	--		--
2014	−.254	*	(.104)	--		--
Number of persons	354,000
Number of person-interviews	1,006,000

Note. SE = standard error.

^*p < 0.05

^**p < 0.01

^***p < 0.001 (2-tailed test).

The within-person level portion of the model, in which each respondent serves as their own control, reveals that the activation of Secure Communities is associated with a significant increase in violent victimization (b = .151, two-tailed p < .05). For the 287(g) jail enforcement and task force policies, the within-person slopes (b = .339 and b = .305) also are positive and relatively large, but the standard errors are large as well, so the estimates are not statistically significant (two-tailed p = .185 and p = .158, respectively). Therefore, in the pooled sample of all racial/ethnic groups, there is no sufficient evidence to say that these policies are associated with adverse victimization outcomes. The evidence, however, is sufficient for rejecting the position that these policies are important for reducing crime. In comparison, the within-person slope for anti-detainer policies (b = .049) is smallest in magnitude among all the policy variables examined, and the estimate is statistically insignificant (two-tailed p = .806). Thus, there is no evidence that the activation of anti-detainer policies has any evident effect on violent victimization risk. The results from the two-way fixed effects specification, displayed in Table 3, support the same conclusions, indicating that the activation of Secure Communities increased personal victimization risk, while the activation of 287(g) and anti-detainer policies did not have a significant effect on whether NCVS respondents experienced a victimization.

Table 3.

Fixed Effects Model Estimated Effects of County Immigration Enforcement Policies on Violent Victimization, 2005–2014

	Fixed effects model
Characteristics	b		(SE)
County policy variables
Secure Communities activated	.118	*	(.052)
287(g) jail enforcement agreement	.378		(.243)
287(g) task force agreement	.393		(.251)
Anti-detainer policy activated	.045		(.098)
Control variables
Individual-level variables
Age	−.081		(.057)
Divorced	.037		(.108)
Separated	.788	***	(.106)
Never married	−.126		(.124)
Education	.014		(.023)
Employed	.092		(.072)
Household income	.004		(.005)
Homeowner	−.032		(.120)
Length of residence	.004		(.004)
Neighborhood-level variables
Population density	.469		(.998)
SES disadvantage	−.389		(.253)
% Black	.015		(.028)
% Latino	−.053		(.030)
% Asian and Pacific Islander	.108		(.078)
% other race/ethnicity	−.238		(.216)
Race entropy	−.733		(1.383)
% foreign born	−.020		(.022)
% ages 18–34	.028	**	(.009)
% divorced/separated	−.018		(.011)
% moved into units <10 years	−.008		(.007)
% vacant housing	.014	*	(.007)
County context
% unemployed	.006		(.012)
Police force size	.709		(.445)
Police expenditures	−.003		(.002)
Survey administration variables
Survey reference time	.217	***	(.013)
Time in sample	−.133	***	(.019)
Year of interview
2006	.057		(.127)
2007	−.016		(.141)
2008	−.059		(.161)
2009	−.205		(.183)
2010	−.263		(.209)
2011	−.267		(.232)
2012	−.295		(.249)
2013	−.524		(.269)
2014	−.553		(.292)
Number of persons	3,800 a
Number of person-interviews	15,500 a

Note. SE = standard error.

^*p < 0.05

^**p < 0.01

^***p < 0.001 (2-tailed test).

^aPersons showing no variability in victimization status across time were not contributing information to the fixed effects analysis, so the sample size of the “effective analysis sample” was smaller.

4.2 |. Race/Ethnicity-Specific Analyses

As discussed earlier, the immigration enforcement policies assessed in the study may have different effects on victimization risk among Latinos, non-Latino Whites, and non-Latino Blacks. To evaluate this, we re-estimated the models reported in Tables 2 and 3 separately for these three racial-ethnic groups. The results are displayed in Table 4. We again see that the two modeling strategies yielded remarkably equivalent results and support the same conclusions.

Table 4.

Race/Ethnicity-Specific Analyses of the Relationships Between County Immigration Enforcement Policies and Violent Victimization, 2005–2014

	Model 1: Longitudinal hybrid model					Model 2: Fixed effects model
	Between-person difference		Within-person change
Characteristics	b	(SE)	b		(SE)	b		(SE)
Panel A: Latinos
Secure Communities activated	−.063	(.183)	.536	*	(.257)	.629	***	(.136)
287(g) jail enforcement agreement	.158	(.129)	.323		(.521)	−.297		(.250)
287(g) task force agreement	.267	(.233)	.826	*	(.415)	.754	**	(.265)
Anti-detainer policy activated	.029	(.209)	.134		(.450)	.368		(.195)
Number of persons		49,500				550
Number of person-interviews		131,000				2,100
Panel B: Non-Latino Whites
Secure Communities activated	.070	(.087)	.136		(.114)	.142		(.094)
287(g) jail enforcement agreement	−.171	(.178)	.548		(.391)	.291		(.177)
287(g) task force agreement	.259	(.137)	.047		(.355)	.027		(.179)
Anti-detainer policy activated	.224	(.130)	−.085		(.265)	−.044		(.148)
Number of persons		237,000				2,600
Number of person-interviews		725,000				10,500
Panel B: Non-Latino Blacks
Secure Communities activated	.184	(.166)	.013		(.216)	−.058		(.130)
287(g) jail enforcement agreement	.043	(.154)	−.575		(.470)	−.578		(.330)
287(g) task force agreement	.368	(.291)	.551		(.473)	.261		(.421)
Anti-detainer policy activated	−.062	(.282)	.333		(.313)	.656		(.581)
Number of persons	36,500					500
Number of person-interviews	97,500					2,000

Note. SE = standard error.

^*p < 0.05

^**p < 0.01

^***p < 0.001 (2-tailed test).

All models included the control variables used in the full-sample analyses (coefficients omitted from the table).

Focusing on the fixed effects models in Table 4, the findings indicate that the activation of the studied policies were not significantly related to victimization risk for non-Latino White and Black respondents. However, the activation of the Secure Communities program and 287(g) task force agreements was found to significantly increase the risk of violent victimization among Latinos. The estimated fixed effects slope of the Secure Communities program (b = .629, two-tailed p < .001) indicates that the program increased violent victimization for Latinos substantially. For a hypothetical “average” Latino respondent (i.e., someone with the mean characteristics of the Latino sample), the probability of experiencing violent victimization in a 6-month window is estimated to have increased by approximately 86% (5.2 per 1,000 to 9.6 per 1,000) when the Secure Communities program was active compared to when it was not active. Meanwhile, the impact of the activation of 287(g) task force agreements on violent victimization among Latinos also was substantial (b = .754, two-tailed p < .01). For a hypothetical “average” Latino respondent, the probability of experiencing violent victimization in a 6-month window is estimated to have increased by approximately 111% (from 5.2 per 1,000 to 10.9 per 1,000) when a 287(g) task force agreement was in place. In contrast to these patterns, the 287(g) jail agreement and anti-detainer policies were found to be inconsequential for Latinos’ experiences with violent victimization: their fixed effects coefficients were relatively small in size and statistically insignificant at the 5% level.

4.3 |. Robustness Analyses

To further explore the robustness of the reported findings, we conducted two additional sets of analyses. First, we re-estimated the models presented above separately for serious and minor violence using multiple definitions that have been applied to NCVS data by BJS.¹⁴ The results revealed that the reported findings hold for both serious and minor violence—that is, victimization risk among Latinos significantly increased with the implementation of the Secure Communities program and 287(g) task force agreements, whereas the 287(g) jail agreements and anti-detainer policies showed no significant effect for any of the racial/ethnic groups studied.

Second, although we presented evidence that supports the plausibility of the parallel trends assumption required for unbiased estimation of the panel models reported above, to assess the robustness of the findings to an alternative approach we replicated the study using a matched DID analysis. This approach has been identified as a useful method when pre-treatment trends in outcomes between treated and untreated cases could be unequal (see Gertler et al., 2011; Imai, Kim, and Wang, 2021).

We implemented this supplementary analysis by first using propensity score matching (Stuart and Rubin, 2008) to match NCVS respondents residing in counties where a given policy of interest was activated (Secure Communities, 287(g), or anti-detainer) with other respondents who had similar characteristics but did not experience policy activation. To do so, we used the observed county characteristics (percent unemployed, police force size, police expenditures, Census region, immigrant destination type), census tract characteristics (percent foreign born, population density, and all other tract variables listed in Table 1), and person/household-level covariates (age, gender, marital status, and all other person/household variables listed in Table 1) to estimate each person’s propensity score of living in a county in which a given policy was activated in the study period. For all analyses, we checked for common support, which showed sufficient overlap in the estimated propensity scores between treated and control observations. The matched samples were created using 1:1 nearest neighbor matching within calipers at 0.25 times the standard deviation of the logit of the estimated propensity scores, as recommended by Rosenbaum and Rubin (1985), though we found that the results were not sensitive to caliper sizes (e.g., 0.2 times the standard deviation produced equivalent results) or whether the matching allowed replacement. We checked balance diagnostics after matching, which showed that the covariate distributions were similar between the matched treated and control group samples.¹⁵

Using the matched samples, we then re-estimated fixed effects models of victimization risk that parallel the specifications shown in Tables 3 and 4. These supplementary analyses revealed that, although the sample sizes became smaller because of matching (i.e., unmatched cases were dropped from the analyses), the results were consistent with those reported above. The matched DID analyses provided additional evidence that the activation of Secure Communities and 287(g) Task Force programs significantly increased violent victimization among Latinos, while having no significant effect for non-Latino Whites and non-Latino Blacks. Additionally, these supplementary analyses affirmed the reported findings that 287(g) jail models and anti-detainer policies were not significantly associated with violent victimization risk for any of the racial-ethnic groups examined during the study period.

5 |. DISCUSSION AND CONCLUSION

In this paper, we described the expansion of Federal-local immigration enforcement policies in the United States and the challenges associated with their evaluation. Our study builds on past research to propose a new data source—the area-identified NCVS data containing information on individual, neighborhood, and county predictors of violent victimization—for evaluating the varied policy approaches. In this section, we summarize the overall findings and assess the implications of the findings against the current research and policy backdrop. To inform future research in this area, we also identify gaps and areas of focus to lay a foundation for additional research on immigration enforcement and crime.

5.1 |. Major Findings and Implications for Current Policy Environment

As contemporary U.S. immigration enforcement is characterized by governments activating many policies in a relatively brief period of time, it necessitates careful evaluation of this complicated policy configuration. Our literature review puts Secure Communities and 287(g) programs in the “intensified interior immigration enforcement” category, for which several studies have found no significant change in police recorded crime rates following the activation of these policies (Forrester and Nowrasteh, 2018; Miles and Cox, 2014; Treyger et al., 2014; Wong, 2012). The anti-detainer policies are in the “integrationist” category, for which studies have ranged from finding mostly “no impact on” to “some reduction of” police recorded crime rates (e.g., Ascherio, 2022; Gonzalez O’Brien, Collingwood, and El-Khatib, 2019; Hausman, 2020; Kubrin and Barto, 2020; Lyons et al., 2013; Males, 2017; Wong, 2017). We build on these studies by reassessing the policies with victimization survey data that includes crimes not reported to the police and permits separate analyses for different racial-ethnic groups.

Anti-detainer policies

Our use of victimization survey data replicated the null findings reported in most previous studies of anti-detainer policies. In doing so, our findings strengthen the observation by Gonzales O’Brien et al. (2019) and Kubrin and Barto (2020) that the idea that anti-detainer policies contribute to crime has no empirical basis (also see Lyons et al., 2013; Males, 2017; Miles and Cox, 2014; Treyger et al., 2014). Nonetheless, the prevalence of anti-detainer programs was modest during our study period, and in the timeframe used in most other studies. It is possible that different patterns would emerge during the more recent period of substantial growth in the activation of anti-detainer policies that occurred after 2016 (Center for Migration Studies, 2022). Researchers should continue to monitor that possibility.

287(g) and Secure Communities

Our research illuminates the importance of considering the different forms in which 287(g) activation manifested during the first decade of its existence. Consistent with previous research (Forrester and Nowrasteh, 2018), our results for 287(g) jail programs showed no significant association with violent victimization risk, a pattern that is evident for Latinos, Whites, and Blacks. However, our findings revealed that 287(g) task force programs significantly increased violent victimization among Latinos, whereas they had no discernible impact among the other groups studied. We observed a similar pattern for the activation of Secure Communities. Thus, our study offers crucial evidence that intensified interior immigration enforcement policies, in the form of 287(g) task force agreements and the Secure Communities program, not only fail to reduce violent victimization, as proponents claim, but they also increased violent victimization among Latinos.

The configuration of findings that emerge from the NCVS data—that Secure Communities and 287(g) task force programs increase violence risk among Latinos, but not other groups—support the plausibility of arguments advanced by many legal scholars and social scientists who have anticipated that 287(g) and Secure Communities programs provide no discernible crime-protective benefits while also bringing adverse consequences for Latinos. The findings lend support to the notion that these programs fostered a “devolution of discretion” (Stumpf, 2015) and net-widening that eroded trust of police in the communities they targeted, putting Latino residents at a higher risk for victimization (Martínez-Schuldt and Martínez, 2021; Nguyen and Gill, 2016). Yet, as we noted earlier, from a theoretical vantage point the activation of Secure Communities and 287(g) task force programs could increase crime for other reasons as well. For example, by facilitating the detention and deportation of large numbers of immigrants from mixed families, these programs may promote significant economic hardships and family disruption that can increase the risk of violence. The data assembled for our study cannot discern between these potential explanations. Although we account for changes in respondent employment status and annual household income, these measures may not be sufficiently precise to capture changes in economic hardships that could increase violent victimizations among minorities (e.g., Lauritsen and Heimer, 2010). Additionally, while the NCVS contains a time-varying measure of respondent marital status, this indicator is unlikely to capture dynamic family separations that frequently occur amid removals facilitated by 287(g) and Secure Communities (Capps et al., 2015). Finally, while the NCVS data used in our study can shed some light on how local immigration policies impact victim crime reporting,¹⁶ the absence of indicators of immigrant status limit the utility of that assessment, and the data do not contain the measures of respondent perceptions of law enforcement or other dimensions of the criminal justice system that would be needed to assess whether 287(g) task force agreements and Secure Communities increased violent victimization risk among Latinos due to changes in legal cynicism. An important need for future research is to explore the mechanisms that may account for the impact of variation in local immigration policies on individual differences in violence risk.

Policy Implications

Whatever the underlying mechanisms, the results of our study support the decision of the Obama administration to discontinue 287(g) task force programs on December 31, 2012 (Kolker, 2021) and President Biden’s executive order that effectively ended the Secure Communities program on January 20, 2021. It is important to recognize, however, that we have seen Secure Communities ended before (November 2014) to only be reinstated a few years later (January 2017). Further, new forms of the 287(g) program emerged to replace the task force model, and participation by local law enforcement agencies has grown considerably over the past several years (Kolker, 2021; Pham, 2018). While these new iterations of 287(g) address some of the earlier concerns emphasized by critics, the program continues to be justified in part on grounds that it is “a tremendous benefit to public safety” (ICE, 2022b) without systematic evidence that this is the case. As we elaborate below, additional research is needed that assesses the impact of newer Federal-local immigration enforcement partnerships and policies on crime. Absent such evidence, we encourage policymakers to consider (1) that the research findings to date provide no evidence that contemporary Federal-local immigration enforcement partnerships have reduced Americans’ exposure to crime, and (2) that our results indicate that such programs may increase violent victimization risk among Latinos.

5.2 |. Additional Avenues for Research

It is vital that the U.S. government supports additional research that assesses the costs and consequences of contemporary immigration policies. Subsequent to the period reviewed in our study, the number of local agencies participating in the 287(g) programs has nearly doubled, increasing to 140 agencies in 24 states (ICE, 2022a). During the same time, the number of jurisdictions that have adopted anti-detainer policies also have grown rapidly (Avila et al., 2018; Center for Immigration Studies, 2022), prompting more states and localities to pass legislation or local policies to either support or curtail anti-detainer efforts. In the first half of 2017 alone, for example, more than 120 state bills were introduced on the issue of sanctuary jurisdictions and immigration detainers (NCSL, 2019). Because of the ongoing intensity of the immigration enforcement debate, it is important to consider evidence from the NCVS and other sources for these continued policies as newer data become available.

Our findings also point to additional areas of interest for future research. First, it would be valuable to extend the policy variables described in this paper to measure how policies are implemented or resisted in practice by measuring the number of police or ICE officers deployed, the volume of detainers issued or refused, the number of ICE arrests, and the prevalence of deportations across jurisdictions (see, e.g., Chand, 2020; Hausman, 2020; Kirksey and Sattin-Bajaj, 2021; Pedroza, 2019). We followed the lead of most previous studies in focusing on policy activation, but from previous research we have learned that, notwithstanding the same clauses of 287(g) agreements or similarly named anti-detainer policies, local communities apply these policies to varying degrees and with varying success (e.g., Arthur, 2018; Coon, 2017; Moinester, 2018; Varsanyi et al., 2012). Therefore, whether the varying degree of enforcement may help explain any observed change in victimization risk is a natural question to extend our study.

Second, future studies of immigration (and other) policy impacts should consider the role of spatial dynamics. The existing research on immigration enforcement (the present study included) has had a local focus: investigating policy effects for people who live in the locality without exploring potential biases associated with the spatial dependence of crime or potential policy spillover effects (e.g., whether effects of an immigration enforcement policy are mainly bounded locally, or whether one locality’s adoption of a certain policy may affect crime risks in another neighboring locality). While previous county-level studies of crime (Baller, Anselin, Messner, and Deane, 2001), simulation research on multilevel data structures (Xu, 2014), and the comprehensive empirical specifications employed in our study suggest that any bias associated with omitting spatial effects in our study would be minimal, emerging evidence from policy analyses suggest that neglecting spatial dynamics in difference-in-differences designs may lead to underestimates of treatment effects (Butts, 2021; Delgado and Florax, 2015), especially when there are compelling reasons to expect spillover effects. Recent research by Kang and Song (2022) suggests that this may be the case with local immigration policy. Their county-level analysis of UCR crime rates indicated that while Secure Communities activation did not reduce crime in “focal counties,” it was associated with a reduction in police-recorded crime when Secure Communities was also activated in adjacent counties, which could reflect deterrence or greater underreporting of crime. It is unclear whether this pattern would hold in analyses that also account for the spatial dependence of crime or potential spatial effects of other variables, but it underscores the importance of considering spatial dynamics in future studies of the impact of immigration (and other) policies on crime. Although the underlying methodology to support such efforts remains “in its infancy,” promising approaches have been suggested for integrating spatial effects in the most common quasi-experimental designs used to test policy effects on crime (see Kolak and Anselin, 2020: 31). Future research that integrates the emerging insights may prove valuable.

Finally, our research underscores the utility of future research integrating victimization survey data to assess the impacts of immigration policy, and policy effects more generally. Previous crime policy research has relied almost exclusively on data from the UCR—recently replaced by the National Incident-Based Reporting System (NIBRS)—which may not provide a complete picture of underlying policy impacts. For our study, the use of the NCVS as an alternative proved valuable because it enabled us to examine policy impacts separately for the three largest racial-ethnic groups in America, which would not be feasible with the UCR data available for the study period. The move to NIBRS should help address this specific limitation of the traditional UCR (Strom and Smith, 2017), but researchers should be mindful of other potential pitfalls in relying exclusively on police-based crime data collections to assess policy impacts. Specifically, while our analysis of NCVS data for 2005–2014 did not uncover systematic evidence that 287(g), Secure Communities, and anti-detainer policies significantly affected victim crime reporting, it remains possible that these policies influenced the probability of crime reporting for groups most directly targeted by them (e.g., the foreign born). Policy researchers should consider that selected policies may affect crime reporting in ways that undermine the utility of applying data based only on crimes known to the police to assess policies. Additionally, research suggests that the UCR exhibited non-random missing data that limited the samples available for assessing the impact of policies across jurisdictions and that the non-random missing pattern may also impact the validity and reliability of estimates for crimes reported to the police (Delang et al., 2022; Maltz and Targonski, 2002). This evidence does not negate the utility of using the UCR data to assess policy impacts, but it does suggests that researchers who use the UCR (or now NIBRS) data to assess policy impacts would be wise to consider the implications of community differences in crime reporting and missing data in drawing conclusions and to consider alternative data sources—such as victimization and offending surveys—to supplement the evidence base available to policymakers.

Appendix

Appendix A. Coding of Control Variables

Individual characteristics
Age	In years
Male	1=yes; 0=no
White (as reference)	1=Non-Latino White; 0=no
Latino	1=Latino (any race); 0=no
Black	1=Non-Latino Black; 0=no
Other nonwhite	1=Non-Latino other nonwhite group; 0=no
Divorced	1=yes; 0=no
Separated	1=yes; 0=no
Never married	1=yes; 0=no
Education	Level of education (0 to 22)
Employed	1=employed during the last 6 months; 0=no. We coded a person as employed only if the job lasted “two consecutive weeks or more,” which was the majority of employed persons, although the results were not sensitive to this requirement.
Household income	Level of household income (1 to 14). To address the issue of missing income, the BJS used a hot deck imputation method to impute income (Berzofsky et al. 2014). As a sensitivity analysis, we also re-estimated all models with the income variable removed and used the victims’ employment status and education to measure the victims’ income-earning abilities. The results were the same across these model specifications.
Homeowner	1=respondent/family owned the home; 0=no
Length of residence	In years
Neighborhood characteristics
Population density	Logged population per square mile
Central city neighborhood	1=yes; 0=no
SES disadvantage	Composite disadvantage index
% Black	Percentage of tract population that is non-Latino Black
% Latino	Percentage of tract population that is Latino
% Asian and Pacific Islander	Percentage of tract population that is non-Latino Asian and Pacific Islander
% other race/ethnicity	Percentage of tract population that is non-Latino other race
Race entropy score	A tract’s entropy score is $E=\sum _{i=1}^5({\prod }_i)\ln [1/{\prod }_i]$ where ${\prod }_i$ refers to a particular racial/ethnic group’s proportion of the tract population. The five racial/ethnic groups are Latinos, non-Latino Whites, Blacks, Asians and Pacific Islanders, and other races.
% foreign born	Percentage of tract population that is foreign born
% ages 18–34	Percentage of tract population aged 18–34 years
% divorced/separated	Percentage of tract population that is divorced or separated
% moved into units <10 years	Percentage of households that moved into units less than 10 years ago
% vacant housing	Percentage of vacant housing units
County context
% unemployed	Percentage of county population that is unemployed
Southern border with Mexico	1=yes; 0=no
Northern border with Canada	1=yes; 0=no
Traditional immigrant county	1=yes; 0=no
New immigrant county	1=yes; 0=no
Police force size	Number of full-time sworn police officers per 1,000 population
Police expenditures	Police expenditures per capita (inflation-adjusted 2010 dollars)
South	1=yes; 0=no
Midwest	1=yes; 0=no
West	1=yes; 0=no
Survey administration variables
Survey reference time	Reference period (1=1 month, 2=2 months, …, 6=6 months)
Time in sample	1=first interview, 2=second interview, …, 7=7th interview
Year of interview	Year dummies (2006 through 2014; 2005 as the reference group)

Appendix B. Maps of Different Types of Counties