1.1 Federal-Local Cooperation on Immigration Enforcement

The Criminal Alien Program (CAP) has been referred to as a “jail check” program with predecessor programs that date back to 1988 (see Reference Cantor, Noferi and MartínezCantor et al. 2015; Reference PatrisiaMacías-Rojas 2016). Under CAP, federal immigration officers screen inmates in federal, state, and local prisons and jails to identify noncitizens who may be removable. After the screening and interview process, CAP officers may issue a request for notification to state or local law enforcement agencies asking to be notified prior to the identified individual's release from the facility so that ICE may assume custody. CAP officers may also issue an “immigration detainer” to have the local facility hold the noncitizen for an additional 48 hours beyond his or her scheduled release date, which facilitates the custody transfer process (see Reference KandelKandel 2016).

Another program known as Secure Communities moved CAP to a universal and automated screening system. Secure Communities achieved this goal by electronically linking local jails to databases maintained by the Department of Homeland Security (DHS) and the Federal Bureau of Investigations (FBI), respectively. When a state or local law enforcement agency arrests and books individuals, the agency sends their fingerprints to the FBI to check against FBI's criminal databases. Secure Communities requires the FBI to send those fingerprints to the DHS to be compared against DHS' biometric identification database, which contains information on removable and potentially removable noncitizens (see Reference Miles and CoxMiles and Cox 2014). Secure Communities began its rollout in 2008 and was activated across all counties in the United States in 2013. However, the Obama administration terminated the program in 2014 and replaced it with the Priority Enforcement Program, which set out enforcement priorities that focused on the removal of serious criminal offenders and recent border crossers (U.S. DHS 2016). According to the Migration Policy Institute, the 2014 enforcement priorities effectively protected 87 percent of unauthorized immigrants nationwide from removal (Reference Capps, Chishti, Gelatt, Bolter and SotoCapps et al. 2018). In 2017, President Trump reactivated Secure Communities through an executive order.

Another key federal-local arrangement that implicates local jails in immigration enforcement is known as the 287(g) program. The program name refers to section 287(g) of the Immigration and Nationality Act. Congress enacted section 287(g) to allow federal immigration enforcement to directly delegate certain immigration enforcement functions to state and local government agencies (U.S. DHS 2010). The law requires that this delegation be executed through a Memorandum of Agreement (MOA) between ICE and the local law enforcement agency. A local jurisdiction that has entered into a 287(g) agreement may employ a jail enforcement model, a task force model, or a combination of the two.

Under the jail enforcement model, deputized local law enforcement officers may work in state and local correctional facilities to identify and process charged or convicted individuals for removal. Under the task force model, deputized local law enforcement officers identify and process removable aliens in community settings during the course of their regular duties as patrol officers, detectives, or criminal investigators (U.S. DHS 2010: 20). The 287(g) program was curtailed during the Obama administration in response to concerns over inconsistent practices and racial profiling, but the program has been revitalized under the Trump administration (Reference Rosenberg and LevinsonRosenberg and Levinson 2017).

1.2 Intergovernmental Service Agreements

There is yet another way that local law enforcement agencies have become deeply intertwined with federal immigration enforcement. Under the Immigration and Nationality Act (8 U.S.C. § 1103(a)(11)(A)), the DHS has authority to contract with state or local government agencies “for necessary clothing, medical care, necessary guard hire, and the housing, care, and security of persons detained by [ICE]….” Over the years, a substantial number of local jurisdictions have entered into contracts with ICE to confine individuals who have been apprehended by immigration authorities.

According to a government report, in fiscal year 2016, ICE owned and operated only five out of hundreds of detention facilities used to hold immigrant detainees (Office of Inspector General 2018). ICE secures the remainder by contracting with private prison companies, the U.S. Marshal Service, or state and local governments. These agreements with state and local governments, known as Intergovernmental Service Agreements (IGSAs), require ICE to pay the local government a per diem for each bed space that it rents out. Some local governments further contract with private prison companies to operate their jails.Footnote ⁵ The IGSAs have come under scrutiny and criticism for allowing ICE to avoid the public bidding process and to operate without standard operating procedures required of other federal awards (Office of Inspector General 2018: 5).

As shown in Figure 1, our analysis of the data on immigration detention released by ICE in November of 2017 (see National Immigrant Justice Center 2017) suggests that the estimated average daily population (ADP) of detainees in local jails started to increase in fiscal year 2010 before beginning to decline in fiscal year 2012. Figure 1 also illustrates a sharp downturn between fiscal years 2014 and 2015 when arrests and removals declined substantially with implementation of the Obama administration's new enforcement priorities in 2014 (see Reference Capps, Chishti, Gelatt, Bolter and SotoCapps et al. 2018: 4). However, Figure 1 shows a rapid uptick after fiscal year 2015. Given that the data do not allow us to identify local jails with precision, Figure 1 represents only a rough approximation—and an underestimate—of the actual ADP of detainees in local jails.Footnote ⁶ Nonetheless, the general trend depicted in Figure 1 underscores the continuing and renewed importance of local jails in maintaining the immigration detention system as it exists today.

Figure 1. Estimated Total ADP of ICE Detainees in Jails in the United States, Fiscal Years 2009–2018. Source: Authors' calculations using detention data obtained by the Immigrant Legal Resource Center from ICE (National Immigrant Justice Center 2017). Note: A fiscal year runs from October 1st of a given year to September 30th of the next year. Fiscal year 2018 estimate is based on partial-year data provided by ICE that ended on November 6, 2017.

What factors predict whether a county will hold immigrant detainees in its local jails? We turn to the longstanding literature on carceral expansion in the United States to develop a theoretical framework for addressing this question.

2.1 Economic Perspective

Studies show that incarceration and prison facilities are concentrated in economically disadvantaged communities (Reference Sampson and LoefflerSampson and Loeffler 2010; Reference ThorpeThorpe 2015). Much research has been devoted to understanding how economic conditions might be linked to incarceration beyond their contributions to criminal offending. Under one prominent strand of social control theory, incarceration is the dominant group's way of controlling the poor or “surplus labor” (for a review, see Reference Western, Kleykamp and RosenfeldWestern et al. 2006). Accordingly, many studies on the economic predictors of incarceration have examined the effects of unemployment (and other related measures such as poverty, wages, and economic inequality). Much of this research has focused on the relationship between labor market conditions in urban areas and mass incarceration. More recently, however, a growing number of scholars have advanced compelling theoretical and empirical reasons to explore this relationship in nonmetropolitan settings as well (Reference Eason, Zucker and WildemanEason et al. 2017; Reference SimesSimes 2018).

Prison siting studies have also examined how economic factors shape prison sitings. The “demand” for prisons in small, rural towns may be fueled by the assumption that prisons can revitalize local economies and promote job growth (Reference Lobao and AduaLobao and Adua 2011; Reference MauerMauer 2006; Reference SchlosserSchlosser 1998). For example, according to Reference LynchMona Lynch's (2009: 133) case study of prison siting decisions in Arizona, in the eyes of local officials, “prisons have foremost become economic development opportunities in both the private and public sectors.” Consistent with this account, studies find that prisons built since the 1970s are disproportionately concentrated in rural, economically disadvantaged areas (Reference EasonEason 2010, Reference Eason2017). By contrast, findings are mixed on the question of whether prison building does in fact produce economic benefits to the host communities (Reference EasonEason 2017; Reference Glasmeier and FarriganGlasmeier and Farrigan 2007; Reference Hooks, Mosher, Genter, Rotolo and LobaoHooks et al. 2010; Reference Hoyman and WeinbergHoyman and Weinberg 2006).

Economic factors could play a role in county decisions to hold ICE detainees in two ways. First, pervasive in media reports are accounts of small towns in financial trouble looking to revitalize their local economy by signing contracts with the federal government to hold ICE detainees. The following report about counties in Arizona succinctly summarizes this common narrative: “Rural towns and counties have eagerly embraced the arrival of immigrant prisoners for the attendant economic benefits, including tax revenues and jobs…. The expanding prison populations have allowed small towns to carry budget surpluses in a state that has otherwise been pummeled by the recession” (Reference KirkhamKirkham 2017). Such reports are neither new nor limited to communities in Arizona; rather, they date back at least a decade and focus on many different localities experiencing economic downturns or financial distress (see, e.g., Reference BernsteinBernstein 2008).

Second, economic incentives to contract with ICE may not be sufficient to participate in immigration detention for counties that lack the basic infrastructure or bed space. At least in some jurisdictions, criminal justice reforms intended to reduce the incarcerated population had the effect of reducing both the prison and jail populations (Reference Garland, Hogan, Wodahl, Hass, Stohr and LambertGarland et al. 2014). Media accounts have noted that the opening up of jail bed spaces produced new pressures or opportunities that enabled struggling counties to become involved in immigration detention (Associated Press 2017; Reference Pickoff-White and SmallPickoff-White and Small 2017). Immigrant detainees could function as a reserve supply of confinable people for struggling facilities looking to fill their jail space.

Taken together, the foregoing discussion suggests that worsening local economic conditions in times of growing bed space in local jails will increase the likelihood that counties participate in immigration detention. We evaluate this expectation in our regression analysis by examining the interaction effect between county unemployment rates and measures of unused or “excess” bed space associated with the criminal jail population.

2.2 Race and Politics Perspective

Another stream of criminal incarceration research investigates the importance of racial dynamics in explaining carceral expansion. Reference EasonJohn Eason (2010), in his study of rural towns, finds that towns with higher percentages of Black and Hispanic populations are more likely to receive a new prison. Other studies suggest that relative size of a community's Black population predicts incarceration rates (Reference Campbell, Vogel and WilliamsCampbell et al. 2015; Reference SmithSmith 2004). The racial threat hypothesis offers one possible explanation for the relationship between the size of the minority population and incarceration rate.

The central claim of the racial threat hypothesis is that whites view growing minority populations as a threat to their political and cultural dominance, which in turns leads to the community's increasing reliance on coercive social control mechanisms, including criminal punishment and incarceration (Reference BlalockBlalock 1967). The relationship between minority population size and criminal incarceration may not be linear, however. Reference Keen and JacobsKeen and Jacobs (2009) show that after the relative size of the minority population reaches a critical threshold, the positive association between minority presence and incarceration reverses. Increasing minority population from already-high levels affords the minority group enough political and economic clout to weaken punitive policies that affect them disproportionately (see also Reference ArvanitesArvanites 2014).

Political factors are distinct, but related, predictors of incarceration. Republicans have traditionally campaigned on law-and-order platforms, which generally favor criminal justice policies that lengthen sentences and expand prison capacity. Likewise, Republican voters are generally more likely to support punitive responses to crime. Accordingly, most studies find that measures of Republican Party strength in communities are significantly and positively associated with those communities' prison incarceration or admission rates (Reference Beckett and WesternBeckett and Western 2001; Reference Jacobs and CarmichaelJacobs and Carmichael 2001; Reference Jacobs and HelmsJacobs and Helms 1996; Reference SmithSmith 2004). Other scholars have argued that the effects of Republican Party strength on incarceration vary over time and by area of the country (Reference Campbell, Vogel and WilliamsCampbell et al. 2015).

Studies tend to investigate racial dynamics alongside political ideology on the theory that incarceration is driven by politics that racialize crime and target minorities for punishment. Reference Keen and JacobsKeen and Jacobs (2009), for example, explore the hypothesis that African Americans should face higher incarceration rates than whites in areas where Republican attempts to link crime with concerns about a “dangerous racial underclass” have been successful.

Although an emerging empirical literature has begun to explore the nature and consequences of immigration detention (see, e.g., Reference Eagly, Shafer and WalleyEagly et al. 2018; Reference Ryo and PeacockRyo and Peacock 2018, Reference Ryo and Peacock2019), no empirical study to date has examined how racial and political dynamics might be affecting decisions of local officials to participate in immigration detention. Local officials are responsible for decisions about whether to sign an IGSA and whether to continue confining ICE detainees in local jails. These local officials are likely influenced by antiimmigrant or proimmigrant sentiments that prevail in their communities and among their constituents.

More concretely, we might expect that the greater the anti-immigrant sentiment, the greater a community's willingness to contract with ICE. Further, studies show that anti-immigrant sentiments are more prevalent in places undergoing influxes of immigrants that trigger heightened perceptions of threat, and when salient political rhetoric and nativism reinforce that sense of threat (Reference Campbell, Wong and CitrinCampbell et al. 2006; Reference HopkinsHopkins 2010). Most relevant to the current study, Reference Creek and YoderCreek and Yoder (2012) find that a growing Latino population and Republican governorship are significant and positive predictors of whether a state signs a 287(g) agreement (see also Reference WongWong 2012). Moreover, Reference Provine, Varsanyi, Lewis and DeckerProvine et al. (2016) find in their study of local law enforcement agencies that a key predictor of agency practices relating to immigrants is the political ideology of local voters. These findings reflect the increasing political polarization on immigration issues, with Democrats generally favoring integrationist approaches and Republicans favoring restrictionist approaches (Reference Hajnal and RiveraHajnal and Rivera 2014).Footnote ⁹

We expect that increases in the county-level Latino population (up to a certain level) and Republican Party strength will increase the likelihood of counties participating in immigration detention. We focus on the Latino population, as existing research suggests that growth and geographic dispersion of the Latino population played a key role in prompting militarized and restrictive enforcement measures in the United States (see, e.g., Reference Massey and PrenMassey and Pren 2012). The majority of the population at risk of immigration apprehension, detention, and removal are Mexicans and Central Americans. For example, in 2001, the first year for which the Yearbook of Immigration Statistics published information on immigration detention by nationality, individuals from Mexico constituted 50 percent of the detainee population, followed by individuals from Guatemala, Honduras, and El Salvador (together, the Northern Triangle region, 16 percent) (U.S. Department of Justice 2003: 235). In 2017, 47 percent of the detainee population originated from the Northern Triangle region, followed by Mexico at 38 percent (U.S. DHS 2019: 9).

3.1 Data

This study draws on several major datasets. The primary data set, the IOB Data compiled by the Vera Institute of Justice, contains county-level information on prison and jail populations over time across the United States. The IOB Data also contain county-level demographic variables such as the county population size, race, and gender of the resident population, and whether a county is urban or rural. As detailed documentation on the IOB Data is available elsewhere (Vera Institute of Justice 2018), we focus our description only on those components of the data directly relevant to the current study.

The IOB Data rely on a number of different sources, including the BJS Census of Jails and the BJS Annual Survey of Jails. The Census of Jails is conducted every five to eight years and contains information on all jails in the United States. The Annual Survey of Jails, fielded almost every year since 1985 except for the census years, is used to estimate the annual total U.S. jail population. The Annual Survey of Jails contains information on approximately one-third of all jails across the country, including nearly all the largest jails and a representative sample of hundreds of others. Together, these two sources aim to provide a consistent annual series of the total number of individuals incarcerated in local jails. Over 3000 counties or county equivalents are included in the IOB Data.Footnote ¹⁰

We compiled and merged into the IOB Data two data sets that relate to major federal-local immigration enforcement partnerships. First, we collected and hand-coded information on whether and when a county entered into a 287(g) agreement with ICE (287g Data). We relied on two primary sources in compiling the 287g Data: (1) archived ICE web pages that show individual 287(g) agreements across all local jurisdictions in the United States, and (2) a list that ICE produced under the Freedom of Information Act on “all 287(g) memoranda of agreements/understanding (MOA) established with ICE since the program was established.”Footnote ¹¹ Our 287g Data contain local jurisdiction names, as well as the start and end dates, if applicable, of individual MOAs signed. Second, we compiled information on Secure Communities (Secure Communities Data) from the ICE “FOIA Library” available on ICE's website (U.S. DHS 2014). The Secure Communities Data contain information on the specific date on which Secure Communities was activated in each county in the United States.

The analysis draws on three additional data sources. First, we used the Bureau of Labor Statistics (BLS) Local Area Unemployment Statistics (https://www.bls.gov/lau/) to capture annual average percent unemployed for each county (BLS Data). The BLS makes the data available on its website for the years 1990 through 2017 (BLS 2018).Footnote ¹² Second, we used data from the Voting and Elections Collection (CQ Press 2018) to create measures of political partisanship (Election Data). The Election Data contain county-level voting results on presidential elections held between 1980 and 2016. Third, to complete the missing information on the crime variables found in the IOB Data (specifically, the IOB Data are missing the years 1993 and 2015 on the crime variables), we used the Uniform Crime Reporting data (UCR Data) found on the Inter-university Consortium for Political and Social Science database (Reference KaplanKaplan 2018).

The data for our regression analysis contain 29,941 county years pertaining to 1252 unique counties.Footnote ¹³ A total of 33 counties had missing data on one or more predictor variables. Because these counties comprised less than 1 percent of all counties, we excluded them using listwise deletion.

3.2 Measures

Table A1 contains detailed descriptions of all of the measures that we discuss below. Our descriptive analysis focuses on Counties with ICE Detainees, which is the total number of counties that held ICE detainees in local jails in any given year. We examine these annual total counts by a number of basic, and relatively stable, demographic, geographic, and political characteristics at the county level. Region indicates the area of the country where the counties are located: (1) Midwest, (2) Northeast, (3) South, and (4) West. Population Size is the total number of county residents (of all ages). Urbanicity distinguishes the counties along an urban–rural spectrum and has four categories: (1) urban, (2) suburban, (3) medium and small metro, and (4) rural (for details on the urban–rural classification scheme, see Reference Kang-Brown and SubramanianKang-Brown and Subramanian 2017: 8).

Political Partisanship captures the counties' voting results in presidential elections for the time period included in our descriptive analysis, 1983–2013. During this time period, some counties remained Republican strongholds, others remained Democrat strongholds, and yet others shifted between Republican and Democrat. To capture these variations categorically in a parsimonious way, we constructed Political Partisanship by first dividing the time period into two groups (1983–1997 and 1998–2013) using 1997 as the midpoint.Footnote ¹⁴ For any given county, if the plurality vote was Republican (Democrat) for the majority of the years in a given time period, we classified that county as Republican (Democrat) for that time period. Using this coding process, we classified each county into the following four categories: (1) Democrat–Democrat, (2) Democrat–Republican, (3) Republican–Democrat, and (4) Republican–Republican.

Our regression analysis uses the following time-varying measures. The dependent variable, Held ICE Detainees, indicates whether a county held any ICE detainee in any given year. Our regression analysis includes the following key predictor variables. Unemployment Rate is the percent unemployed in a given county's labor force. Excess Bed Space is a measure of unused bed space for jail, pretrial, and other jail populations, respectively. As coded in the IOB Data, these three populations do not overlap and are defined as follows: Jail population refers to convicted criminal offenders serving time in local jails. Pretrial population refers to criminal defendants held pretrial in local jails. Other jail population refers to individuals held in local jails for all other federal, state or other local law enforcement and corrections agencies excluding ICE.

For each of these jail populations, we separately measured the annual total number of unused or “excess” bed space in each county, z _it, by applying the following formula:

(1) $z_{\mathrm{it}}=1-\left(\frac{x_{\mathrm{it}}-\min \left(x_i\right)}{\max \left(x_i\right)-\min \left(x_i\right)}\right),$

where x _it is ADP (of jail population, pretrial population, and other jail population, respectively)Footnote ¹⁵ of county i at time t; min(x _i) and max(x _i) are the observed minimum and maximum ADPs, respectively, across all years within county i. Subtracting the observed minimum ADP from a given year's ADP and dividing the result by the difference between the observed maximum ADP and the observed minimum ADP normalizes the ADP values to a range of 0 and 1. Subtracting these normalized values from 1 produces an estimate of how many bed spaces could have been filled by the jail population, pretrial population, and other jail population, respectively, but were not. Each of the three Excess Bed Space variables associated with each type of jail population is thus a relative measure (relative to the observed maximum within each county) of unused bed space.Footnote ¹⁶

Percent Latino is the percent of the county population aged between 15 and 64 that is Latino. Following previous studies on racial threat, we include a quadratic term for the Latino population in our regression model to test whether the effect of Latino population on counties participating in immigration detention is curvilinear; if so, the Percent Latino measure should have a significant positive effect, whereas the quadratic term should have a significant negative effect.Footnote ¹⁷

Republican indicates whether the plurality vote of a given county was Republican in any given presidential election. This coding stays constant for all intervening years between presidential elections.

We use the following control variables. Secure Communities indicates whether Secure Communities was activated in a given county for any given year. To be coded as 1 on Secure Communities, the program had to have been active in the community for at least one day during a given year. Active 287g indicates whether a given county had an active 287(g) agreement with ICE during a given year. We coded this variable as 1 if the county had an active 287(g) agreement for at least one day during a given year. If only a city within a county had an active 287(g) agreement, we did not code the county in which the city is located as having an active 287(g) agreement, as our unit of analysis is county years.

In addition, our regression models also include two key crime-related control variables. Violent Crime Rate is the total number of violent offenses per 1000 county residents between the ages of 15 and 64. Property Crime Rate is the total number of property offenses per 1000 county residents between the ages of 15 and 64.

3.3 Analytical Strategy

Our analysis unfolds in two stages. First, to understand temporal and spatial patterns of detention proliferation, we consider changes over time (1983–2013) in the total number of counties across the United States that held immigrant detainees in their local jails. We conduct this analysis descriptively by examining a series of graphs that show temporal changes by a key set of basic county characteristics. For this descriptive analysis, we focused on county characteristics that are more or less constant over time in order to ensure the relative stability of counties' membership in various categories shown in each of the graphs.

Second, to identify the factors associated with the likelihood of a county holding immigrant detainees in its local jails, we conduct a series of logistic regressions using conditional maximum likelihood. We examine the effects of various independent variables of interest on the log-odds of counties holding immigrant detainees using year and county fixed-effects models. Fixed-effects models cannot estimate the effect of predictors that are constant over time. However, fixed-effects models allow us to address omitted variable bias. Specifically, our two-way fixed-effects models control for (1) any changes over time that impact all counties, and (2) any observed and unobserved time-invariant heterogeneity (e.g., region, rural–urban distinction) between counties. For example, the year-specific fixed effect absorbs any changes in law, policies, or political environment at the national level that might affect the likelihood of counties getting involved in immigration detention. Additionally, our county-specific fixed effect absorbs all stable observable or unobservable differences across counties not represented by the variables in our model.

As a robustness check, we reestimated our fixed-effects models using lagged covariates (lagged by one year in relation to the dependent variable) to test the possibility that the relationships at issue operate on longer time gaps than what we have estimated. The results were substantively and statistically similar to the results obtained by models that do not use lagged variables, as shown in Table A2. Note that fixed-effects logistic regression analysis uses only within-county variation. For counties that did not hold any ICE detainees and counties that always held ICE detainees during our analysis period, there is no within-county variation in the outcome variable, and the fixed-effects logistic regression drops these counties. We replicated our logistic regression analysis that specifies random intercepts for counties (with year fixed effects), which allows us to include these dropped counties, and we found results that are similar to the results from the fixed-effects models. The results of the random-effects models are shown in Table A2. Our discussion focuses on the fixed-effects models, however, as Hausman tests comparing the fixed-effects models against the random-effects models favored fixed-effects models.Footnote ¹⁸

Our analyses focus on the period 1983–2013 because this is the period for which necessary data are available. Because the IOB Data contain information on ICE detainees starting in 1983, our descriptive analysis begins with 1983. However, our regression analysis begins with the year 1990 because measurement on one of our key independent variables—the size of the Latino population at the county level—is available beginning only in 1990. The period of observation ends in 2013 because this is the most recent year for which the BJS Census of Jails—the census that covers all counties, as discussed above—has released its data.Footnote ¹⁹

We do not analyze changes in the number of ICE detainees as an outcome variable in this study. The measure of ICE detainee population size, ADP, found in the IOB data is distinct from the number of detainees admitted into facilities in any given year. It is also distinct from the total number of detainees held in facilities in any given year. As noted earlier, ADP represents the sum of daily populations across all local jails in a given county, divided by 365 days. This means that changes in the ADP of ICE detainees in any given county reflect some combination of changes in the number of new admissions and changes in the composition of detainees who continued to be detained from the previous year. Insofar as these compositions change over time within any given county, it is difficult to reliably estimate the relationship between various predictors of interest and annual changes in the detainee population level using the ADP.

Additional challenges unique to the immigration detention context further complicate the use of ADP as an outcome variable. To make the ADP comparable across counties, it needs to be standardized by some underlying rate of individuals “at risk” of detention. However, we cannot reliably measure the relevant “at risk” population (e.g., noncitizens subject to mandatory detention, undocumented immigrants) because these types of data are not available, particularly at the county level for the range of years that we examine. Moreover, because many ICE detainees experience interfacility transfers during their detention stay, the population “at risk” of detention who reside in a given county ultimately may not be detained in that county.

4.1 Descriptive Analysis

Our descriptive analysis maps changes—over time and across various county characteristics—in the total number of counties that held immigrant detainees in their local jails. Figure 2 shows the total number of county jails holding immigrant detainees between 1983 and 2013. Figure 2 reveals an increasing number of counties over time, steadily rising from 128 counties in 1983 to 727 counties in 2013—nearly a sixfold increase.

Figure 2. Total Number of Counties with ICE Detainees by Year, 1983–2013. [Color figure can be viewed at wileyonlinelibrary.com] Source: IOB Data. Note: The line corresponds to a linear prediction plot. The years are calendar years.

The most substantial rise occurred in the South, followed by the Midwest and the West, as Figure 3 illustrates. Compared to these other regions, the Northeast remained relatively flat over the years. Within regions, a large share of the growth was concentrated within a relatively small number of states. By 2013, approximately half of all counties holding ICE detainees were in just ten states: Texas (107 counties holding ICE detainees), Georgia (45), North Carolina (35), California (34), Florida (28), Kentucky (25), Virginia (24), Wisconsin (22), Indiana (22), and Colorado (20). While states like Texas, California, and Florida have been longstanding destinations for international migrants, many of the states with high incidence of counties holding ICE detainees represent the so-called “new destinations” without the same tradition of immigration (see, e.g., Reference MarrowMarrow 2005).

Figure 3. Total Number of Counties with ICE Detainees by Region, 1983–2013. [Color figure can be viewed at wileyonlinelibrary.com] Source: IOB Data. Note: The line corresponds to a linear prediction plot. The years are calendar years.

Figure 4 shows that the most notable increase in the number of counties participating in immigration detention took place in small counties with population sizes between 10,000 and 249,999. By contrast, the smallest counties (less than 10,000) and the largest counties (1 million or more) experienced the least growth. Mapping the trend by urban–rural spectrum provides us with an additional perspective. Figure 5 shows that while the number of counties holding immigrant detainees remained relatively flat in urban areas, rural, and medium/small metro areas experienced a steady growth, followed by suburban areas. These trends mirror the findings from research on criminal incarceration documenting the expanding use of jails in small counties (defined as counties with less than 250,000 people) and rural areas (nonmetropolitan areas remote from major population centers) since the 1970s (Reference Kang-Brown and SubramanianKang-Brown and Subramanian 2017).

Figure 4. Total Number of Counties with ICE Detainees by Population Size, 1983–2013. [Color figure can be viewed at wileyonlinelibrary.com] Source: IOB Data. Note: The line corresponds to a linear prediction plot. The years are calendar years. For the population size categories, we adopted the National Center for Health Statistics' classification scheme (National Center for Health Statistics 2014).

Figure 5. Total Number of Counties with ICE Detainees by Urbanicity, 1983–2013. [Color figure can be viewed at wileyonlinelibrary.com] Source: IOB Data. Note: The line corresponds to a linear prediction plot. The years are calendar years. For details on how the urbanicity categories are coded, see Reference Kang-Brown and SubramanianKang-Brown and Subramanian (2017).

Finally, we examine the relationship between political partisanship and involvement of counties in immigration detention. As discussed earlier, we divided the years between 1983 and 2013 into two periods, with 1997 as the midpoint. We then identified each county's modal plurality vote (Democrat or Republican) in presidential elections across the pre-1997 years, and across the years 1997 and after, respectively. Our coding produced four categories of Democrat–Democrat, Democrat–Republican, Republican–Democrat, and Republican–Republican. Republican–Republican counties, for example, are counties that voted Republican between 1983 1997 as well as between 1997 and 2013. Figure 6 shows a steep increase in the number of counties participating in immigration detention in the Republican–Republican category.

Figure 6. Total Number of Counties with ICE Detainees by Political Partisanship, 1983–2013. [Color figure can be viewed at wileyonlinelibrary.com] Source: IOB Data and Election Data. Note: The line represents a linear prediction plot. The years are calendar years. Political partisanship is coded based on voting results from presidential elections. For details on how the political partisanship categories are coded, see Table A1.

In sum, the analyses show a steady rise in the number of counties engaged in immigration detention in the past three decades. Localities that became involved in immigration detention during this period are not randomly placed. Rather, the most notable growth occurred in relatively small, nonurban areas that are largely Republican and located in the South.

4.2 Multivariate Analysis

Our two-way fixed-effects logistic regression analysis investigates the predictors of counties holding ICE detainees in local jails. Table 1 presents basic descriptive statistics on the variables used in our logistic regression models. The results of the regression analysis are presented in Table 2. A positive (negative) coefficient indicates that an increase in a given predictor variable x increases (decreases) the log-odds of counties holding ICE detainees. We interpret the magnitude of the estimated coefficients by converting them to odds ratios, which represent the estimated factor by which an independent variable predicts the odds of a given outcome. We do not present marginal effects estimates because doing so would assume that all time-invariant county-level effects are equal to zero (see Reference DowdNorton and Dowd 2018: 875).

Table 1. Means/Proportions for Variables Used in Fixed-Effects Logistic Regression, 1990–2013

N = 29,941 county-years pertaining to 1252 unique counties.

Table 2. Fixed-Effects Logistic Regression Results Predicting Counties Holding ICE Detainees

Note: N = 29,941 county-years. All models include year and county fixed effects. Standard errors in parentheses.

^*p < .05; ^**p < .01; ^***p < .001 (two-tailed tests).

Because Violent Crime Rate and Property Crime Rate are highly correlated (r = .689), we first estimated the models using Violent Crime Rate. We then replaced Violent Crime Rate with Property Crime Rate in the same set of models. Our findings are robust to this alternative specification. To save space, we present the results from the models containing Violent Crime Rate (alternative model results are available upon request). The regression models are nested to first include only the key predictor variables of interest (baseline models), to which we add the control variables discussed earlier (full models). Given that the inclusion of these control variables to the baseline models (Models 1a, 2a, and 3a) leaves the estimated effects of the key predictor variables unchanged, our discussion focuses on the full model results (Models 1b, 2b, and 3b).

Model 1b includes Excess Bed Space—Jail Population, the measure of unused bed space for the criminal jail population, and its interaction with Unemployment Rate, along with Percent Latino, Percent Latino Squared, and Republican. The interaction term indicates that the positive effect of increasing Unemployment Rate on the likelihood of holding ICE detainees is greater when values of Excess Bed Space are high. This suggests that excess bed space positively moderates the relationship between local unemployment rate and the likelihood of counties holding ICE detainees. Akaike information criterion, Bayesian information criteria, and likelihood ratio test statistics consistently indicate that the model with the interaction term, compared to an alternate model without the interaction term, is a better fitting model.

Model 1b also shows that each percent increase in Latino population within a county predicts a 29 percent increase ((exp [0.255] – 1) x 100) in the odds of that county holding ICE detainees. However, the negative sign of the quadratic term for Percent Latino indicates that this relationship attenuates as Percent Latino increases. Finally, Model 1b shows that political partisanship of voters at the county level is a significant predictor. Specifically, changes in voting results that favor a Democrat candidate in one presidential election to a Republican candidate in the next presidential election increase the odds of counties holding ICE detainees by about 50 percent ((exp [0.401] – 1) x 100). Participation in the Secure Communities and 287(g) program increase the odds of counties holding ICE detainees, suggesting that immigration detention at the local level may be part and parcel of broader local-federal cooperative arrangements on immigration enforcement more generally.

Models 2a and 2b, respectively, are identical to Models 1a and 1b, except that the former set of models replace Excess Bed Space—Jail Population with Excess Bed Space—Pretrial Population. Similar to Model 1b, Model 2b shows that the interaction between Excess Bed Space—Pretrial Population and Unemployment Rate is positive and significant. Likewise, the results for Percent Latino, Percent Latino Squared, and Republican, are generally the same as in Model 1b. Similar to what we found in Models 1a and 2b, a variety of model fit tests indicate that the model with the interaction term is a better fitting model than the one without the interaction term.

Finally, Models 3a and 3b, respectively, are identical to Models 1a and 1b, except that the former set of models replace Excess Bed Space—Jail Population with Excess Bed Space—Other Population. Unlike in Models 1b and 2b, Model 3b shows that the interaction between Excess Bed Space—Other Population and Unemployment Rates is not statistically distinct from zero (p < .05). However, the results for Percent Latino, Percent Latino Squared, and Republican, are generally the same as those that we obtained in Models 1b and 2b.

Our regression analysis produced a notable set of patterns controlling for county fixed effects, year fixed effects, local policies related to immigration enforcement, and local crime rates. First, increases in excess bed space for the jail and pretrial populations appear to amplify the positive effect of unemployment on the likelihood of counties participating in immigration detention. Second, an increase in the Latino population consistently predicts an increase in the probability of counties participating in immigration detention, though this relationship attenuates as the Latino population continues to rise. Third, Republican Party strength in presidential elections also predicts a higher likelihood of participating in immigration detention. We conclude by considering the implications of these findings below.