Beyond the Border and Into the Heartland: Spatial Patterning of U.S. Immigration Detention

Abstract

The expansion of U.S. immigration enforcement from the borders into the interior of the country and the fivefold increase in immigration detentions and deportations since 1995 raise important questions about how the enforcement of immigration law is spatially patterned across American communities. Focusing on the practice of immigration detention, the present study analyzes the records of all 717,160 noncitizens detained by Immigration and Customs Enforcement (ICE) in 2008 and 2009—a period when interior enforcement was at its peak—to estimate states’ detention rates and examine geographic variation in detention outcomes, net of individual characteristics. Findings reveal substantial state heterogeneity in immigration detention rates, which range from approximately 350 detentions per 100,000 noncitizens in Connecticut to more than 6,700 detentions per 100,000 noncitizens in Wyoming. After detainment, individuals’ detention outcomes are geographically stratified, especially for detainees eligible for pretrial release. These disparities indicate the important role that geography plays in shaping individuals’ chances of experiencing immigration detention and deportation.

Appendix

Identifying Unique Detention Episodes in the 1998–2010 Data Set

Each detention event an individual experiences represents one row in the data set, and the data are ordered chronologically from the first detention event to the last. As such, I identify detention episodes by linking subsequent rows that shared the same initial book-in date, gender, and country of citizenship. A hypothetical example illustrates this process: Guatemalan female first detained on September 1, 2006, transferred six days later to another facility, and held for 10 days before being deported. The first row of data for this woman would have an initial book-in date of 09/01/06, a book-in date of 09/01/06, and would indicate that the individual was transferred on 09/07/06 to X facility. This row would also include information about the initial detention facility and apprehension date and location. The next row would indicate that the individual was initially detained on 09/01/06, the book-in date would be 09/07/06, and other fields would indicate that she was deported on 09/17/06. This row would also include the same information as the row above with respect to the apprehension date and location.

Process of Merging the Data Sets

Because the data sets do not share a unique identifier, when merging the data, I construct a person signature and an episode signature for each detention episode consisting of the shared person characteristics in the two data sets (i.e., gender and country of origin) and the shared detention characteristic (i.e., the date and location of each stay in the detention system and the reason for release). I use Python to merge records that matched in both their person and episode signatures. For nonunique combinations of signatures, I create an algorithm that sorts through the state in which the apprehension occurred and the charge that led to detention. If at least one of these fields is consistent across the duplicate combinations of signatures, then they are merged. For example, if each data set includes two sets of records pertaining to Mexican men who were detained in the same facilities on the same dates and were released for the same reason, both sets of records are merged only if either the state in which the apprehension occurred and/or the charge are consistent across both records. So if each of these individuals was first apprehended in Texas, the duplicate pairs of records are matched at random with the records in the other data set even if the charge varies. However, if both the charge and the state of apprehension³¹ vary across both data sets, none of the records are merged. As such, in the case of duplicate combinations of signatures, I cannot be confident that specific individuals in each data set are correctly matched. However, I am confident that the individual characteristics used in the analysis are attributed to the correct state of apprehension, which is the central unit of analysis. Fifty-four percent of the merged records have a unique person and episode signature combination, and a total of 12,544 detention episodes (1.6 % of all episodes) are not merged because the state in which the individual was apprehended and the charge that led to detention varies across the records. The remainder of unmerged records do not have a matching record in the other data set. Of these unmerged records, approximately 90 % are from the 1998–2010 data set.

Geocoding Landmarks

There are more than 9,000 unique landmarks in the data set for 2008 and 2009. These landmarks fit into six main categories: (1) city and county names; (2) the jail or prison facility from which an individual was first transferred into immigration custody; (3) specific enforcement programs names, such as 287(g), the Law Enforcement Agency Response Unit (LEAR), Fugitive Alien, and Criminal Alien Program (CAP) programs; (4) ports of entry; (5) landmarks located along the Mexico–U.S. border, including monument numbers, mountain ranges, bridges, ghost towns, creeks, and passes; and (6) highway and road intersections. I first geocode these landmarks using Google’s Geocoding API in R. I then independently geocode each landmark by hand twice and cross-validate these results. In this second stage, I identify both the state in which the landmark is located and whether the landmark is located along the U.S. border. For cases in which the landmark (such as a county name) could be assigned to several states, I cross-reference the landmark with the initial states of detention for all individuals apprehended at that landmark and assign the landmark to the state(s) in which the detentions occurred.

To identify apprehensions of border crossers, I use the apprehension landmark field, the field specifying the original date of entry into the United States, and counts of apprehensions at that location. Figure 1 details this process. Common border landmarks include ports of entry; border monuments; ghost towns along the border; and geographical areas, such as mountain ranges and valleys, where the U.S. Customs and Border Protection has a strong presence according to local newspaper articles and the social media feeds of border patrol stations and border vigilantes. For landmarks located within 100 miles of the border that are not clearly linked to border patrol activity, counts of apprehensions are examined in conjunction with newspaper and social media analysis. For example, there is extensive news coverage of unauthorized immigrants attempting to avoid checkpoints in Falfurrias, Texas, 70 miles north of the border, by crossing through nearby private ranches by foot. Therefore, when the apprehension landmark field includes the name of a ranch located near the Falfurrias checkpoint that had multiple apprehensions (often >50) between 2008 and 2009, I code it as a border landmark.

Excluded Detention Outcomes

Excluded from the regression analyses are detentions that ended with the following outcomes: (1) order of supervision; (2) transfer to U.S. Marshall; (3) parole; (4) withdrawal; (5) escape; (6) death; (7) transition to an alternative to detention; and (8) transfer to the Office of Refugee Resettlement. Together, records with these outcomes account for 4 % of all merged detention episodes in 2008 and 2009. Orders of Supervision account for approximately one-half of these records; U.S. Marshal transfers account for an additional 40 %; parole, 7 %; and withdrawal, 2.5 %. The remaining outcomes combined account for less than 1 % of the detentions in these years. Orders of supervision occur when a detainee is released after a final order of removal. In these cases, ICE has not met the time limits imposed for deporting the individual, often because of challenges in acquiring the needed documentation from the detainee’s home country. A detainee is turned over to the U.S. Marshal Service typically when a criminal case against the individual is outstanding or the individual is needed as a material witness in a criminal case. Parole occurs when an individual is granted temporary permission to enter the United States. Parole is most frequently granted when a detainee shows demonstrated need to enter the United States for medical or humanitarian reasons. Withdrawal occurs when an individual’s request to enter the United States is allowed to be withdrawn. Orders of supervisions and transfers to the U.S. Marshal are excluded from the analysis because they are triggered by factors that cannot reasonably be tied to one’s apprehension location. Paroles and withdrawals are excluded because they apply only to people attempting to enter the country. The other outcomes are excluded because of how infrequently they occur during the observation period.

Construction of Detention Capacity Measure

Following Schriro’s (2009) methodology, I estimate detention capacity as the average daily population of detainees in each detention facility from January 1, 2008 to December 31, 2009. These counts as well as data on the noncitizen population are then aggregated to the field office level. There are 24 field offices located in 18 states (ICE n.d.). For California, Texas, and New York, each of which has multiple field offices, I combine detention capacity calculations in order to merge in ACS data on the noncitizen population and create detention rate estimates at the field office level. For example, Texas has four field offices that oversee enforcement across Texas, New Mexico, and Oklahoma. Each field office is responsible for a region of Texas, such as North Texas and Central South Texas. Because it is unclear from the information that ICE provides which cities and communities make up these regions, estimating the noncitizen population in that area is not possible. I therefore combine the average daily population of detainees in each detention facility across all three states, as well as the number of detentions and the noncitizen population estimates of all three states. In this way, I can compare the average detention capacity across all three states to the average detention rate for the three states. New York, on the other hand, has two field offices that oversee enforcement only in New York. New York’s aggregated detention capacity measure is thus the average detention capacity of the state.

I estimate detention capacity at the field office level rather than the state level because field offices manage detention space. Mississippi provides a useful case in point. Mississippi has a large number of immigration apprehensions in this period but zero ICE detention beds. As a result, the New Orleans Field Office, which oversees enforcement in Mississippi, detains individuals from Mississippi in Louisiana, which has a number of large detention centers. Calculating detention space at the state level would fail to capture the extent to which Louisiana’s bed capacity may affect detention decisions for individuals apprehended in Mississippi.

Construction of Interior Enforcement Scale

To create the interior enforcement scale, I draw on data compiled by Leerkes et al. (2012, 2013), making several changes to the data to better fit the analysis needed for this study. First, I extend the data set to all 50 states. Their scale is restricted to the 43 states for which the Pew Hispanic Center has provided estimates of the unauthorized immigrant population. Second, because I am interested in only indirect measures of enforcement (i.e., the presence of restrictive immigration-related laws or enforcement programs), I drop the component of their scale that captures the rate of the estimated unauthorized population in a given state-year arrested through the Secure Communities program. Third, I supplement the data on the presence of state laws regarding unauthorized immigrants’ access to a driver’s license—obtained from the National Conference of State Legislatures database, which starts with laws enacted in 2005—with data from a Congressional Research Service Report (Smith 2005). The result is the inclusion of 13 states that had restrictive ID laws on the books prior to 2005 that remained in effect through the observation period. Fourth, rather than using dummy variables to indicate the presence of a city or county 287(g) or Secure Communities contract, I merge data from the 2007–2009 ACS three-year estimates of the noncitizen population in each city and county and then calculate the share of a state’s noncitizen population living in a jurisdiction with an active 287(g) or Secure Communities contract. Several of the cities and counties that participated in 287(g) and Secure Communities in 2008 and 2009 had small populations and, particularly, small noncitizen populations. A 287(g) or Secure Communities contract would likely have a greater effect on a state’s immigration detention rate in jurisdictions with larger noncitizen populations. The scale now accounts for these potential differences. Last, because data are not available on the percentage of firms in a given state participating in E-Verify—a federal program that allows employers to electronically verify the work eligibility of new hires—for the seven states not included in Leerkes et al.’s (2012, 2013) analyses, I look only at the presence of state laws requiring E-Verify usage.

This scale is thus the sum of the following measures, each of which is standardized. I look only at 2009 because no state in 2009 overturned a policy or stopped participating in an enforcement program that was active in 2008, but multiple states passed policies or initiated enforcement programs in 2009. This scale is plotted against a state’s average higher-bound detention rate from 2008–2009 (see Fig. 5).

• 1. Dummy variable indicating the presence of a state 287(g) contract.

• 2. Dummy variable indicating the presence of a state law requiring employers to participate in E-Verify.

• 3. Dummy variable indicating the presence of a state law restricting access to public benefits for undocumented immigrants residing in the state.

• 4. Dummy variable indicating the presence of a state law restricting access to IDs for undocumented immigrants residing in the state.

• 5. Dummy variable indicating the presence of a state law restricting access to employment for undocumented immigrants residing in the state.

• 6. Share of a state’s noncitizen population residing in a county with a 287(g) contract.

• 7. Share of a state’s noncitizen population residing in a city with a 287(g) contract.

• 8. Share of a state’s noncitizen population residing in a county participating in Secure Communities.

Construction of Sanctuary Ordinance Measure

This measure captures the share of a state’s noncitizen population living in a jurisdiction with a sanctuary ordinance in place prior to the end of 2009. I draw on FAIR (2017) and Seghetti et al. (2006) to identify cities and counties with sanctuary measures in place prior to the end of 2009 and then merge data from the 2007–2009 ACS three-year estimates of the noncitizen population in each of these cities and counties. To avoid double-counting individuals, I exclude from the analysis the sanctuary ordinances passed in Salem, Oregon, and San Francisco, California, because the counties in which these two cities are located also had sanctuary ordinances and are included in the analysis. The measure is thus the sum of noncitizens within a given state living in a jurisdiction with a sanctuary policy over the entire population of noncitizens in that state. By the end of 2009, only 18 states had passed some form of a sanctuary measure.

Table 2 Lower- and higher-bound estimates of state detention rates

Table 3 Descriptive statistics for nonmandatory detentions, 2008–2009

Table 4 Sequential logistic regressions of detention outcomes

Table 5 Logistic regression of pretrial release for nonmandatory detentions