Is Previous Removal From the United States a Marker for High Recidivism Risk? Results From a 9-Year Follow-Up Study of Criminally Involved Unauthorized Immigrants

Abstract

The present study examines the long-term recidivism patterns of a group of unauthorized immigrants identified to be at high risk of recidivism. Using a sample of 517 male unauthorized immigrants, we used three measures of recidivism to assess 9-year rearrest differences between unauthorized immigrants who have and who have not been previously removed from the United States. Results indicate that prior removal was a significant risk marker for recidivism, with previously removed immigrants showing a higher likelihood of rearrest, a greater frequency of rearrest, and a more rapid time-to-first rearrest. While the present study does not establish whether previous removal is a consistent indicator of high recidivism, it suggests that this group of unauthorized immigrants may be worthy of review and policy consideration. Much potential value for law enforcement lies in the sharing of federal immigration records with academics to further study the outcomes of unauthorized immigrants.

Keywords

immigration recidivism illegal immigrants jail populations propensity scores

Particularly since the terrorist attacks of September 11, 2001, the enforcement of civil immigration law within the interior of the United States has been an increasing focus of attention and controversy (Saunders, Lim, & Prosnitz, 2010). Estimates suggest that this population has grown to 11 million, up from an estimated 8 million in 2000 (Passel & Cohn, 2012). What to do about this population has been the subject of considerable controversy. There is, however, comparably less rancor over potential enforcement responses to unauthorized immigrants who are identified within a criminal justice population. Perhaps as a result, this particular subgroup (i.e., criminally involved unauthorized immigrants) has been an increasing focus of federal immigration enforcement efforts over the past decade (U.S. Government Accountability Office [GAO], 2012). Within this subgroup, a high priority has been placed on identifying and removing those who have previously been removed from the country (Morton, 2012).¹

Central to this focus is the expectation that criminally involved unauthorized immigrants with a record of previous removal pose a high relative risk to public safety. To date, however, almost no empirical information has been available to test this expectation. We could identify only a single study (Hickman & Suttorp, 2010) on the topic. This work found a substantial difference in rearrest patterns, in a direction of greater recidivism risk for those with a record of previous removal relative to those without. This study, however, is limited by a relatively short 1-year follow-up period. Previous research with other criminal justice populations suggests that short-term differences in recidivism do not necessarily predict long-term differences (e.g., Huebner, 2009; Johnson, 2004; Maltz, 1984; Ostermann, 2013; Worrall, 2008). The purpose of the current study is to examine whether the considerable differences in rearrest detected over a short period in the Hickman and Suttorp study persist over a substantially longer follow-up period. While rearrest measures cannot be used as an indicator of underlying criminal propensity (Elliott, 1995), empirical analyses based on these data can contribute useful information for informing federal and state/local policy choices around unauthorized immigrants. In particular, analyses of the relationship between previous removal and recidivism can assist federal policy makers in evaluating unauthorized immigrant detention and removal priorities. Moreover, the availability of empirical data about rearrest may also help local jurisdictions in weighing the complex policy choices surrounding the extent and form of their active cooperation with federal authorities in immigration enforcement.

Interior Immigration Law Enforcement

Historically, interior immigration enforcement (i.e., enforcement in areas not immediately adjacent to the U.S. border) has been the purview of the federal government, with local law enforcement agencies largely unconnected to such efforts (Vogel et al., 2009). In recent years, however, the involvement of local law enforcement expanded dramatically with the passage of the Illegal Immigration Reform and Immigration Responsibility Act of 1996. This federal legislation included Section 287(g), which allowed the federal government to essentially “deputize” members of state and local law enforcement agencies to perform certain enforcement roles previously reserved for federal authorities. These include designating specially trained officers to substitute for federal immigration authorities within local jails to identify and seek legal detainers for unauthorized immigrants. The first Section 287(g) agreement was formed in 2002, and by 2009, there were 833 so-called 287(g) local law enforcement officers in 23 states (Office of Inspector General, 2010). Section 287(g) also enabled state and local law enforcement participation in targeted immigration enforcement task force activities, contingent upon the development of formal agreements detailing mutual responsibilities and training requirements.

Secure Communities

In 2008, the federal government also launched “Secure Communities.” This is an effort intended to increase the identification and removal of criminally involved immigrants by tapping into the standard practices of state and local law enforcement. Under Secure Communities, when state and local law enforcement agencies send digitized fingerprint and arrestee information to the Federal Bureau of Investigation (FBI), the standard records check query is also routed through a separate immigration records check. The initiating law enforcement agency then receives two types of information: (a) the standard criminal records information (including “wants and warrants”) and (b) information about whether the individual is an unauthorized immigrant eligible for removal. In the latter case, an alert is also sent to the corresponding U.S. Immigration and Customs Enforcement (ICE) field office. The ICE field office may then request that the law enforcement agency detain the individual for transfer into federal immigration custody for potential detention and removal from the United States (GAO, 2012).

According to a report released by ICE, as of January 22, 2013, Secure Communities has been “activated” in 100% of U.S. local jurisdictions, accounting for 3,181 jurisdictions in all 50 states, the District of Columbia, and all U.S. territories (ICE, 2013a). Between its launch in late October 2008 and 2012, ICE reports that approximately 5% of over 28 million Secure Communities searches resulted in a match with an individual in federal immigration records. The process ultimately led to the removal of 227,590 unauthorized immigrants during this time period (ICE, 2013c).

Unlike Section 287(g) collaborations between local and federal agencies, Secure Communities has been characterized as an information-sharing program between federal agencies (i.e., FBI and Department of Homeland Security) and, therefore, immigration records checks do not require the active consent of the local agencies submitting arrestee fingerprints (ICE, 2012). While Secure Communities has been welcomed in some jurisdictions, in others it has been implemented over the objections of local and state elected officials and vocal public protest (Jordon, 2011).

With all local U.S. jurisdictions now participating directly and indirectly in interior civil immigration enforcement, a number of issues and challenges have arisen in the process. Among them are concerns about collateral consequences for policing in local communities. These include worsening relations between immigrant communities and law enforcement, leading to suppression of overall crime reporting, reduced cooperation with law enforcement investigations, and thwarting of community policing efforts (Davis, Erez, & Avitabile, 2001; Decker, Lewis, Provine, & Varsanyi, 2009; Khashu, 2009; Vogel et al., 2009). In addition, a number of community protests have arisen around the assertion that Secure Communities is merely sweeping up large numbers of low-risk immigrants for minor offenses. Opposition groups often highlight examples of individuals transferred to ICE custody after arrests for public intoxication, unlicensed street ice cream sales, and unpaid parking tickets (Harmon, 2012; Taxin, 2011).

The U.S. Department of Homeland Security, in turn, has sought to clarify that its interior civil enforcement efforts are focused on individuals who pose a noteworthy risk to the community, not on individuals arrested for minor offenses or with a limited criminal history (Office of Inspector General, 2010). In its most recent clarification across all its interior enforcement programs, ICE Director John Morton issued guidance directing agents to target individuals “whose removal promotes public safety, national security, border security, and the integrity of the immigration system” (Morton, 2012).

Previously Removed Unauthorized Immigrants

Among the characteristics identified as top ICE interior enforcement priorities are unauthorized immigrants with a prior record of (or outstanding order of) removal. This group has long been singled out in political and policy discussions as posing a particularly high crime risk, potentially cycling back and forth across the U.S. border engaged in criminal pursuits (Countywide Criminal Justice Coordination Committee [CCJCC], 1997; Raymond, Hickman, & Williams, 2004; Warner, 2005). Through fiscal year 2012, ICE reports that Secure Communities has led to the apprehension and removal of 42,596 unauthorized immigrants who had either been previously removed or had previously evaded final orders of removal/deportation. This represents approximately 19% of all removals of individuals identified through Secure Communities (ICE, 2013c).

Despite the priority placed on identifying and removing this group, there has been almost no empirical assessment of the relative criminal recidivism risk posed by previously removed unauthorized immigrants (Saunders et al., 2010). This is a particularly relevant issue at a time when ICE detention resources are increasingly limited and must be strategically utilized for those who pose the greatest risk (Morton, 2012). Moreover, local governments in some jurisdictions are under growing pressure to refuse immigrant detainer requests through Secure Communities or have already declared their intention to refuse them for individuals arrested for minor offenses (Rubin & Blankstein, 2012).

While we could identify no theoretical perspective that contemplates previously removed immigrants, there is reason to suspect that unauthorized immigrants with a record of prior removal may pose a higher risk of criminal recidivism. Given that prior criminal behavior and failure to comply with federal law can cause a removal, a record of prior removal may simply serve as an indicator of a prior official criminal record. The latter is a very well-established recidivism risk factor (e.g., Cullen & Gendreau, 2000; Latessa, Listwan, & Koetzle, 2013; MacKenzie, 1997). Empirically, though, this expectation remains untested. In our search of the empirical literature, we could find a rapidly growing macro-level literature examining immigrant population concentrations and crime (e.g., Ousey & Kubrin, 2009; Wadsworth, 2010), and many fewer individual-level studies (e.g., Katz, Fox, & White, 2011; Powell, Perreira, & Harris, 2010). We could identify only study containing data distinguishing unauthorized immigrants at the individual level. This study found that “legal” immigrants identified in a criminal justice population did not differ from unauthorized immigrants on the likelihood, frequency, or timing of rearrest 1 year post-jail release (Hickman & Suttorp, 2008). Likewise, we could identify only one study distinguishing previously removed immigrants (Hickman & Suttorp, 2010). While there are a few examples of reports released by federal agencies that provide some descriptive information on the previously removed, none contain a comparison group or report more than simple counts or percentages with a rearrest or conviction.

In the next section, we discuss the sole existing study on previously removed unauthorized immigrants in some length to set the context for the present study.

Los Angeles County Jail Foreign Born Inmates Study

Hickman and Suttorp’s study of previously removed unauthorized immigrants used data from the Los Angeles County Jail Foreign Born Inmates study (Hickman, Suttorp, & Wong, 2005). The latter study focused on male foreign-born jail inmates released between August 4 and September 2, 2002. Over this 30-day period, there were a total of 2,892 foreign-born inmates released from the jail. Within this group, 1,617 (56%) individuals were determined to be unauthorized immigrants. Of this latter group, 517 (32%) were released physically from the jail into the community (as opposed to a release directly into the custody of ICE or another agency).² Of those, 27% had a federal immigration record of at least one prior removal/deportation from the United States. Using these data, Hickman and Suttorp (2010) compared the 1-year rearrest patterns of the community-released unauthorized immigrants with a record of previous removal (n = 139) to the rearrest patterns of those without such an arrest (n = 378).

Defining recidivism as a record of rearrest at any point within 365 days of each individual’s date of jail release, Hickman and Suttorp found a sizable and significant difference in recidivism between these two groups. Specifically, 73% of those with a record of previous removal were rearrested within 1 year, relative to 32% of those without. The former group was also arrested more frequently, with 28% arrested 3 or more times relative to 7%, respectively. These large differences in the simple bivariate comparisons also emerged in multivariate statistical models examining the likelihood, frequency, and timing of rearrest. The two groups began to diverge around 35 days post-jail release, with a record of prior removal associated with more rapid rearrest.

Present Study

Previous studies of recidivism within other types of offender populations have found that short-term differences do not necessarily persist over the longer term (e.g., Johnson, 2004; Maltz, 1984; Ostermann, 2013; Worrall, 2008). The goal of the present study is to assess whether the rearrest differences observed in the Hickman and Suttorp (2010) study persist after a substantially longer follow-up period. The present study utilizes the same baseline data as Hickman and Suttorp (2010) but adds updated arrest records for all 517 individuals covering 9 years post-jail release (2002-2011). In the next section, we discuss the identification of unauthorized immigrants in the original data and the subsample used in the current study, followed by a description of the independent and dependent variables used in the analysis. Finally, we discuss the analytic approach.

Description of Data and Variables

The Los Angeles County Jail Foreign Born Inmates study (Hickman et al., 2005) collected data generated through the standard jail booking procedures for foreign-born inmates. Each arrestee was routinely asked to self-report their place of birth and submitted to digital fingerprinting using Livescan equipment. Those indicating foreign birth or for whom an existing criminal justice record indicated foreign birth were flagged for immigration status determination. (The United States possesses no national database of native-born citizens. There is therefore no practical research method of confirming the native citizenship of the large group of individuals who claim it.) Digital fingerprints were also routinely searched against local, state, and federal databases to reduce the risk of false identification (Raymond et al., 2004).

Immigration status for all foreign-born inmates was determined via direct screening in the Los Angeles County Jail by ICE agents and via immigration records checks for each foreign-born individual conducted by the ICE’s Law Enforcement Support Center (LESC).³ For purposes of the original data collection, ICE agents entered an immigration detainer record for all foreign-born inmates they determined to be unauthorized (via in-person interviews and record checks). When ICE agents were unavailable to interview foreign-born inmates directly (largely due to time constraints), electronic queries were sent by jail staff to the LESC. In response, LESC staff provided electronic replies containing information about the presence or absence of an immigration status record for each individual. If an immigration record could be found in one of multiple federal immigration database checks, LESC provided information about the individual’s current legal status and whether the individual had been removed (or had evaded prior final orders to leave the country) prior to the 2002 jail stay.

Unauthorized immigration status was coded for the Los Angeles County study using (a) these electronic responses from the LESC or (b) the presence of an unauthorized immigration detainer entered by an ICE agent located within the jail (Hickman et al., 2005). In the data, unauthorized status is indicated as a single variable representing both individuals who had originally entered the country illegally and those whose legal permission to be in the country had either expired or had been revoked.

Previous removal status was coded for the original study based on LESC electronic responses indicating the presence of such a record (=1). In six cases, the presence of previous removal status was coded (=1) based on a record in Los Angeles County’s own fingerprint-based criminal records system. The latter contains a flag indicating previous removal for a small number of unauthorized immigrants. This flag was the result of a temporary immigration record-sharing partnership between immigration authorities and Los Angeles County for the purposes of increasing the identification of this group (see Raymond et al., 2004, for a full discussion of this arrangement) . In sum, for all unauthorized immigrants, the data also contained an indicator capturing whether the individual had been previously removed or had evaded final orders of removal (=1) or no such record could be found (=0).

Given the present focus, additional discussion of the potential limitations of the sample is in order. It is reasonable to expect high motivation on the part of unauthorized immigrants to provide false identification information to avoid detection. This identification problem is not unique to the study of immigrant populations. Recidivism research of all sorts that relies on official records suffers from this problem to varying degrees, with an expected bias in the direction of failing to identify those with more extensive criminal histories. Fortunately for the present study, Los Angeles County had developed an integrated electronic fingerprint-based criminal justice records system, making positive identification more likely at the time of 2002 jail booking (Applegate & Chotiner, 1999).⁴ Digitized fingerprints are matched against not only local arrestee and court records (e.g., arrest, conviction, and sentencing information) but also those maintained at the state (maintained by California Department of Justice) and national level (maintained by the FBI’s National Crime Information Center; Raymond et al., 2004). The process improves the accuracy of existing records and reduces the likelihood that individuals can conceal their identities by providing false names (Applegate & Chotiner, 1999). This is only the case, however, for those with an existing record. LESC immigration records checks were made using the individual identifiers provided by Los Angeles County after positive identification.

Of course, it is still likely that some unauthorized immigrants in the sample avoided identification, particularly among those individuals with no prior criminal justice contact. This raises questions about the extent to which undetected unauthorized immigrants differ from those positively identified in the current sample. Among the 1,698 foreign-born inmates released into the community, LESC could identify no matching immigration record for 24%. As the absence of such a match was not accepted by ICE as adequate evidence of unauthorized status, these unmatched inmates were coded as immigration status missing (Hickman et al., 2005). While this likely means the data underestimate the number of unauthorized immigrants, it avoids a larger alternative problem of clouding the interpretation of results by “guessing” at immigration status. Like the Hickman and Suttorp (2010) analysis, those missing immigration status were excluded from the present sample. An analysis (using t tests and chi-square tests) showed that these excluded cases did not differ significantly on available demographic characteristics, criminal history, or recidivism patterns.

Control variables

In addition to immigration status (and previous removal status), the baseline data contained inmate criminal history records, limited demographic information, and data on the circumstances and characteristics of the 2002 jail stay. The current study used the same control variables as the Hickman and Suttorp (2010) study. That studyemployed variables based on what could be obtained from available records in addition to their relevance in prior recidivism research, particularly a prior study of individuals released from the Los Angeles County Jail (Petersilia, Turner, & Fain, 2000).

Demographic characteristics were drawn from the jail’s management information system and included age at the time of 2002 jail release and country of birth. The majority of the sample (62%) self-reported Mexico as their place of birth, with a mix of Central American countries representing the next most commonly reported location (27%). Given this distribution, nationality is captured with a dichotomous variable indicating Mexico (=1) versus any other country (=0). Given the lack of variation in the sample, race and ethnicity information were not included as control variables.

In the present context, controlling for criminal history is important because of the expectation that the existence of an official criminal record could represent a systematic difference between the two groups of interest (i.e., unauthorized immigrants with and without a record of previous removal). For this purpose, relevant criminal history variables refer to the period prior to the 2002 jail stay. Two variables capture criminal history: (a) a frequency variable indicating the number of prior arrests and (b) a dichotomous variable indicating at least one prior conviction for either a misdemeanor or felony offense (=1). Criminal history data were obtained from Los Angeles County’s Consolidated Criminal History Reporting System (CCHRS; based on fingerprints), statewide criminal records from the California Department of Justice, and national criminal history information from the FBI’s National Crime Information Center. These local, state, and national criminal history data were consistently available for a 6-year period prior to the study’s data collection in 2002.

The circumstances of the 2002 jail stay were captured in a series of variables obtained from the jail’s management information system. Several variables captured variation in type and severity of offenses associated with the 2002 jail stay (each stay could be associated with one or multiple offenses). For capturing severity of offense, a dichotomous variable was used indicating whether the most serious charge associated with the jail stay was a felony (=1) versus a misdemeanor or violation (=0). Because individuals can be incarcerated for multiple offenses, the data contained four dichotomous variables capturing whether any one of the charges included at least one person offense (=1), at least one property offense (=1), at least one drug offense (=1), or at least one other type of offense (=1). The latter category includes vehicle/driving-related offenses, weapons charges, parole/probation violations, and the like.

A frequency variable captured the number of days spent in the Los Angeles County jail prior to release (i.e., number of days of incarceration) and another dichotomous indicator assessed whether the individual was released post-conviction (=0) or was not convicted at the time of release (=1). The latter group includes those released from the jail due to trial acquittal, dropped charges, and bond/citation. The presence of high multicollinearity among the control variables was tested for and ruled out using variance inflation factor (VIF) diagnostics.

Recidivism measures

For the current study, the 9-year recidivism variables consist of three measures. Likelihood of rearrest (=1) is an indicator capturing whether criminal arrest records show at least one arrest within 9 years (3,285 days) of the 2002 jail release date. A frequency variable captures the number of discrete arrest incidents recorded over the 9-year period. In other words, an arrest incident can include multiple charges but is counted as only one arrest. Finally, time to rearrest captured the number of days from the day of the 2002 jail release until first rearrest, with a possible range of 1 to 3,285 days for each individual. Table 1 shows a summary of the study characteristics and data sources.

Table 1.

Summary of Study Characteristics and Data Sources.

Fielding period	August 4, 2002-September 2, 2002
Inmate release categories	Released pretrial on citation or bond, case dismissed or acquitted, or post-conviction release after sentence completion
Data used to code unauthorized status	a. Jail records of immigration screenings and/or detainers b. Law Enforcement Support Center narrative response to law enforcement query for immigration status
Data used to code previous removal	Law Enforcement Support Center narrative response to law enforcement query for immigration status
Control variables from official jail records	Demographics, criminal history, current charges, reason for release, and length of stay associated with incarceration during the fielding period
Recidivism data	Criminal records provided by the California Department of Justice through 2011

Statistical Methodology

We used three methods to analyze differences in the 9-year recidivism between the two groups of unauthorized immigrants. First, we examined likelihood of rearrest by using a logistic regression model to assess whether previous removal was predictive of any rearrest. Logistic regression measures the relationship between a dichotomous event (i.e., rearrested vs. not rearrested) and a set of independent variables (Hosmer & Lemeshow, 1999). Second, we used survival analysis to compare previously removed and never removed immigrants on the timing of rearrest. We estimated time to rearrest using a Cox proportional hazards model with right censoring (again, because not all were rearrested by the end of the follow-up period).

Finally, we examined frequency of rearrest using negative binomial regression to model the relationship between previous removal and the number of rearrests incurred during the follow-up period. In this context, a negative binomial model is the more appropriate count regression approach (relative to a Poisson model) because it adjusts for overdispersion, which relaxes the assumption that the mean and the variance of the number of rearrests are equal (Long, 1997), and accommodates count data with several zero values (i.e., those who were not rearrested during the follow-up period).

A sensitivity analysis was performed to test for the potential influence of extreme outliers in the negative binomial analysis. Extreme outliers were defined as individuals with a frequency of rearrest that exceeded the boxplot “inner fence” (calculated as quartile 3 + 1.5 × interquartile range; for example, see Dawson, 2011). Using this method, we identified 17 extreme outliers (3% of the sample), each with 13 or more rearrests in the 9-year follow-up period. Seven of these were among the previously removed and the remaining 10 had no record of prior removal. The sensitivity analysis results indicated the influence of extreme outliers on three control variables but not on the key variable of interest (i.e., previous removal status). Unfortunately, there is no consistent standard in the literature for dealing with extreme outliers under these circumstances (e.g., see Orr & Sackett, 1991). Analysts are generally advised to consider the interpretation of results. Given this, we elected to drop the 17 extreme outliers from the negative binomial analysis reported here (though we will return to them in later analyses). This choice does not impact the findings on previous removal and avoids undue attention to control variables that are disproportionately influenced by less than 3% of the sample. There may be something unique about the 17 extreme outliers producing such a high rate of rearrest. Unfortunately, they are too few and our data are too limited to provide insight on this issue.

As prior removal status is not randomly assigned, there is a likelihood that the groups differ substantially on background characteristics in ways that could impact results. Controlling for potentially relevant differences increases confidence in results but comparisons can be improved in this context through the use of propensity score weighting in multivariate models. Propensity score weighting is a method to adjust for confounding variables that are correlated with both the key predictor variable (prior removal status) and the outcome of interest (arrest; McCaffrey, Ridgeway, & Morral, 2004; Ridgeway, 2006; J. L. Rosenbaum, 1987). Weighting by propensity scores provides a more robust comparison than does a standard regression approach for estimating the average recidivism outcome for a previously removed immigrant who is situated similarly (with respected to observed control variables) to the remaining unauthorized immigrants released from the jail (McCaffrey et al., 2004; Wong & Schonlau, 2013).

To calculate the propensity weights, using the unweighted baseline covariates shown in Table 3, we estimated a logistic regression model that predicts whether an unauthorized immigrant was previously removed. All previously removed immigrants received a weight of 1, while never removed immigrants received a weight equal to the predicted value from the logistic regression divided by 1 minus the predicted value (p / 1 − p; P. Rosenbaum & Rubin, 1984). Table 3 shows that the application of the propensity score weights effectively balances the two groups for each covariate (i.e., after weighting, the two groups do not differ significantly on any of the baseline covariates). In all of the propensity score-weighted models, we incorporated the same covariates to control for additional bias and for variance reduction (Bang & Robins, 2005). We also used robust standard errors to account for the propensity weights in each model.

Results

The goal of this study is to test whether previous removal is a risk marker for recidivism over a long-term follow-up period among unauthorized immigrants identified within a local jail population. First, we present the simple bivariate relationships between rearrest and previously removed status over the 9-year follow-up period. Next, we present the results for the multivariate models that examine these relationships while holding constant other potentially confounding factors.

Bivariate Analyses of Recidivism

Overall, 71% of all 517 unauthorized immigrants were rearrested at least once in the 9-year follow-up period. Among those with a record of previous removal, 91% were rearrested at least once compared with 64% of those never removed. This is a significant and sizable difference in likelihood of rearrest (Pearson χ² = 34.57, df = 1, p < .001). Among the 369 unauthorized immigrants who were rearrested at least once, rearrest occurred more quickly in the follow-up period for those with a record of previous removal. Specifically, over the 9-year (3,285 days) follow-up period, the average number of days until first rearrest among the previously removed was 251.27 (SD = 400.58) compared with 484.03 (SD = 603.21) for those never removed (t = 4.42, p < .001).

Table 2 shows the distribution of rearrests by previously removed status. After 9 years, 62% of the previously removed immigrants had been rearrested 3 or more times compared with 38% of the never removed immigrants. A greater proportion of previously removed also had five or more rearrests relative to those with no record of prior removal (36% vs. 20%, respectively). The never removed group, however, contained a small number of individuals with the highest frequency of rearrest: One had 45 rearrests in the follow-up period, with the next three extreme outliers having 25, 22, and 20 rearrests, respectively. All of the previously removed immigrants had 19 rearrests or fewer.

Table 2.

Number of Rearrests by Previously Removed Status Over the 9-Year Follow-Up Period (N = 517).

	9-year frequency of rearrests
Number of rearrests	Never removed (n = 378)	Previously removed (n = 139)
0	135 (36%)	13 (9%)
1	58 (15%)	14 (10%)
2	42 (11%)	26 (19%)
3	40 (11%)	12 (9%)
4	26 (7%)	24 (17%)
5	13 (3%)	7 (5%)
6	10 (3%)	11 (8%)
7	9 (2%)	8 (6%)
8	14 (4%)	4 (3%)
9	8 (2%)	2 (1%)
10	7 (2%)	7 (5%)
11	2 (1%)	1 (1%)
12+	14 (4%)	10 (7%)
	Pearson χ² = 60.89, df = 11, p < .000***

Note. Percentages may not add to 100 due to rounding.

p < .05. **p < .01. ***p < .001.

The overall mean number of rearrests was 3.3 (SD = 4.25). Previously removed immigrants had an average of 4.5 (SD = 3.9) rearrests, while never removed immigrants had an average of 2.8 (SD = 4.3; t = −4.07, p < .001). The median number of rearrests for each group was 4 and 1, respectively. As discussed previously, 3% of the sample consisted of extreme outliers on this measure, with a rearrest frequency of 13 or greater during the follow-up period. Removal of the 17 extreme outliers impacts the means for both groups slightly but does not eliminate the significant difference between them. Absent the extreme outliers, the mean number of rearrests for the previously removed group (n = 132) is 4.0 (SD = 3.0) and 2.4 (SD = 2.9) for the never removed (n = 368; t = −5.28, p < .001).

The bivariate comparisons revealed that the previously removed were more likely to be rearrested and to be rearrested more often in the follow-up period. These differences, however, may be due to differing background characteristics. To examine this possibility, we conducted multivariate analyses with propensity score weighting to equalize the groups on the observed control variables. As shown in Table 3, the propensity score weighting procedure effectively balanced the two groups on each of the 11 control variables. This can be observed through the non-significant F statistics (computed from the Pearson χ² statistic with the Rao & Scott, 1981, 1984, second-order correction based on design effects).

Table 3.

Baseline Covariates Before and After Propensity Score Weighting (N = 517).

Variable	Previously removed (n = 139)	Not previously removed (n = 378) weighted values		Test of previously removed and not previously removed (weighted)
Variable	Previously removed (n = 139)	Weighted	Unweighted	Test Statistic	p value
Country of birth				F^db = 0.048	.826
Mexico	86%	86%	53%
Other	14%	14%	47%
M (SD) age at release	28.2 (7.8)	27.9 (7.5)	32.4 (9.7)	F(1, 516) = 0.09	.766
M (SD) number of previous arrests	4.7 (4.3)	5.5 (6.0)	2.7 (3.9)	F(1, 516) = 1.17	.230
At least one previous conviction	74%	76%	59%	F^db = 0.163	.687
Most serious charge = felony	48%	49%	31%	F^db = 0.012	.913
At least one person charge	17%	16%	22%	F^db = 0.001	.978
At least one property charge	19%	23%	15%	F^db = 0.312	.577
At least one drug charge	38%	37%	13%	F^db = 0.049	.824
At least one other charge	38%	38%	58%	F^db = 0.002	.967
M (SD) number of days in jail^a	22.2 (30.0)	22.5 (28.6)	21.8 (28.5)	F(1, 516) = 0.01	.936
Release type: Not convicted	54%	54%	47%	F^db = 0.002	.966

Note. F^db = design-based F statistic testing weighted data.

Days in jail was capped at 100 due to extreme outliers in fewer than 5% of the cases.

We used the propensity score weights (and robust standard errors to account for them) when estimating all multivariate models and included the baseline covariates to control for additional bias and for variance reduction (Bang & Robins, 2005).

Multivariate Analyses of Recidivism

In multivariate analyses, we used three different measures of long-term recidivism: likelihood, frequency, and number of days until rearrest. First, we present the results of a logistic regression analysis testing whether the two groups of unauthorized immigrants differed on the likelihood of rearrest, while adjusting for baseline characteristics. As shown in Table 4, the results of this analysis revealed a significant difference between the two groups: The odds of having at least one rearrest during the 9-year follow-up period were 2.62 times higher for those with a record of previous removal, relative to those without. Other baseline characteristics significantly associated with rearrest were a higher number of previous arrests and age (younger immigrants were significantly more likely to have a rearrest than were older immigrants).

Table 4.

Logistic Regression of Rearrest During 9-Year Follow-Up Period (N = 517).

	Odds ratio	Robust SE	z	p > \|z\|	95% confidence limits
Previously removed**	2.62	0.909	2.78	.005	[1.33, 5.17]
Country of birth—Mexico	1.01	0.370	0.03	.977	[0.49, 2.07]
Number of previous arrests***	1.25	0.071	3.87	<.001	[1.11, 1.39]
At least one previous conviction	1.21	0.454	0.51	.607	[0.582, 2.52]
Age at release***	0.932	0.016	−4.00	<.001	[0.901, 0.965]
Most serious charge = felony	0.631	0.228	−1.27	.203	[0.310, 1.28]
At least one person charge	1.40	0.873	0.54	.591	[0.412, 4.75]
At least one property charge	2.39	1.74	1.19	.232	[0.573, 9.95]
At least one drug charge	2.49	1.80	1.26	.207	[0.604, 10.27]
At least one other charge	1.00	0.592	0.01	.995	[0.316, 3.19]
Number of days in jail	1.00	0.005	0.74	.459	[0.994, 1.01]
Release type: Not convicted	1.55	0.485	1.39	.165	[0.836, 2.86]

p < .05. **p < .01. ***p < .001.

Second, we used the Cox proportional hazard regression model (Hosmer & Lemeshow, 1999) to examine the number of days until first rearrest during the 9-year follow-up period (see Table 5). The hazard function represents the risk that an individual is arrested at a point in time, given that no arrest had occurred by that day. Using the same covariates and propensity score weighting as shown in Table 3, the results again indicate a significant difference between the groups. A record of previous removal was associated with fewer days until rearrest. Time until rearrest was also significantly associated with number of previous arrests and age at the time of 2002 jail release.

Table 5.

Survival Analysis of Days Until First Rearrest During 9-Year Follow-Up Period (N = 517).

	Coefficient	Robust SE	z	p > \|z\|	95% confidence limits
Previously removed***	.583	0.158	3.69	.001	[0.274, 0.893]
Country of birth—Mexico	−.003	0.185	−0.01	.988	[−0.366, 0.361]
Number of previous arrests***	.070	0.012	5.57	.001	[0.045, 0.094]
At least one previous conviction	.027	0.180	0.15	.879	[−0.325, 0.380]
Age at release***	−.047	0.010	−4.78	.001	[−0.067, −0.028]
Most serious charge = felony	−.144	0.205	−0.70	.482	[−0.546, 0.258]
At least one person charge	.213	0.221	0.96	.336	[−0.221, 0.647]
At least one property charge	.130	0.223	0.58	.559	[−0.307, 0.567]
At least one drug charge	.465	0.238	1.95	.051	[−0.001, 0.931]
At least one other charge	−.066	0.223	−0.30	.766	[−0.504, 0.371]
Number of days in jail	.001	0.002	0.43	.665	[−0.004, 0.006]
Release type: Not convicted	−.129	0.163	−0.79	.430	[−0.449, 0.191]

p < .05. **p < .01. ***p < .001.

Figure 1 displays the survival function for the two groups of unauthorized immigrants, using weighted data. As the figure shows, the two groups began to diverge early in the follow-up period, with the previously removed showing more immediate and rapid failure.

Figure 1.

Days until first rearrest by previous removal status.

The final multivariate analysis tested whether there was a difference between the two groups in the frequency of rearrest during the 9-year follow-up period. In this analysis, we used negative binomial regression. It adjusts for overdispersion, relaxing the assumption that the mean and the variance of the number of rearrests are equal (McCullagh & Nelder, 1989). As previously discussed, approximately 3% of the sample (n = 17) were extreme outliers in frequency of rearrest, and a sensitivity analysis revealed no differences in results for the key variable of interest. In the final results reported here, these extreme outliers were dropped because of their influence on three control variables; thus, the final model presented in Table 6 describes the relationships for 97% of the sample (n = 500). Results with the full sample of 517 included are available upon request.

Table 6.

Negative Binomial Regression of Frequency of Rearrest During 9-Year Follow-Up Period (n = 500).

	Coefficient	Robust SE	z	p > \|z\|	95% confidence limits
Previously removed**	.298	0.010	2.98	.003	[0.102, 0.494]
Country of birth—Mexico	−.045	0.115	−.039	.698	[−0.269, 0.180]
Number of previous arrests***	.061	0.013	4.66	<.001	[0.036, 0.087]
At least one previous conviction	−.077	0.137	0.56	.574	[−0.192, 0.346]
Age at release***	−.030	0.007	−4.55	<.001	[−0.042, −0.017]
Most serious charge = felony	−.056	0.111	−0.50	.616	[−0.274, 0.163]
At least one person charge	.170	0.154	1.10	.270	[−0.132, .473]
At least one property charge	−.248	0.202	−1.23	.220	[−0.643, 0.148]
At least one drug charge	.128	0.173	0.74	.458	[−0.210, 0.467]
At least one other charge	−.254	0.146	−1.74	.081	[−0.539, 0.032]
Number of days in jail	−.002	0.002	−1.25	.210	[−0.005, 0.001]
Release type: Not convicted	−.240	0.120	−2.00	.045	[−0.474, −0.005]

p < .05. **p < .01. ***p < .001.

As Table 6 shows, previous removal is a significant predictor of frequency of rearrest, while taking into account baseline characteristics. Other baseline characteristics significantly associated with the frequency of rearrest were a higher number of previous arrests and age at the time of release (younger immigrants were more likely to have a higher number of rearrests).

Discussion

The present study is an examination of the long-term recidivism patterns of a group of unauthorized immigrants identified in prior research as posing a significant recidivism risk. Specifically, Hickman and Suttorp (2010) found that a record of previous removal was a distinct risk marker for rearrest among unauthorized immigrants identified within a local criminal justice population. After 1 year, 73% of those with a record of prior removal had been rearrested compared with 32% of those without. Among those with a record of prior removal, 25% were rearrested 3 or more times within 1 year, relative to just 7% of those with no prior removal. Even given these substantially large differences, there were no empirical grounds upon which to anticipate a priori the direction of the long-term results with this population. Indeed, studies of recidivism within other criminal justice populations have found that short-term differences do not necessarily persist over the longer term (e.g., Johnson, 2004; Ostermann, 2013; Worrall, 2008). Indeed, Maltz (1984) argued that the convention of a 1-year follow-up period in recidivism studies is often determined more by research funding limitations rather than by substantive decisions about the population under study.

The current study uses the same baseline data and replicates the methodology used in this previous study (Hickman & Suttorp, 2010) but extends the follow-up period from 1 year to 9 years. Overall, the long-term findings are remarkably consistent with the short-term findings. Previous removal remained a significant risk marker for recidivism, regardless of whether rearrest was measured in terms of likelihood, time-to-first rearrest, or frequency of rearrest. In comparison with similarly situated unauthorized immigrants with no prior record of removal, those with a record of removal were more likely to be rearrested; only 9% of previously removed immigrants had no record of rearrest during the 9-year follow-up period, compared with 36% of the never removed immigrants. Furthermore, previously removed immigrants were rearrested more frequently, with a mean number of 4.5 new arrests over the 9-year period compared with 2.9 for the never removed unauthorized immigrants. The median number for each group was 4 and 1, respectively. The time to rearrest findings were consistent as well, with previously removed immigrants showing a much more rapid first rearrest. This latter finding, however, is not surprising given the 1-year findings that 73% of the previously removed had already been rearrested in the first year.

In discussing the relevance of these results, we first discuss limits in their interpretation and then, within that context, offer comments on their potential utility. The study was not able to examine and compare types of rearrest. Therefore, if the two groups varied on types of rearrest charges (e.g., severity), these differences could not be detected. As with all studies relying on official criminal justice records, the present study uses data that capture at least as much about the behavior and operation of criminal justice systems and victim reporting practices as it does individual arrestees (Elliott, 1995; Geerken, 1994; Maltz, 1984). Moreover, we had no means of describing or controlling for the varying amount of time individuals may have been reincarcerated over the entire 9-year follow-up period. For example, the first or any subsequent arrest during the 9-year period could have resulted in a lengthy incarceration, making that individual “ineligible” to recidivate again during the follow-up period. Of course, this is the case for any long-term study of recidivism that relies on rearrest measures.

In the context of immigration research, however, potential detection, detention, and removal by immigration authorities present an additional risk that may also make individuals ineligible to recidivate. Indeed, ICE apprehensions of previously removed immigrants within the United States have increased during the study’s follow-up period (GAO, 2011). We did not have access to follow-up immigration records from ICE or LESC for the sample and therefore cannot describe or control for the influence of immigration enforcement on these results. The increasing emphasis on immigration enforcement within criminal justice populations, however, could lead one to speculate that there was an increased risk of ICE apprehension and removal for the specific individuals in our sample. All criminal history factors being equal, ICE enforcement priorities suggest that the previously removed are at greatest risk of post-rearrest ICE detention and removal (relative to those LESC deems unauthorized but never previously removed).

Despite this increased enforcement, over the 9-year follow-up period of the present study (2003-2011), the previously removed were rearrested more frequently. All things being equal, if risk of detention and removal increased more for the previously removed, we would expect a lower frequency of rearrest relative to those with no record of previous removal. One potential explanation is that all factors were not equal. It may be that the previously removed are simply exceptionally high-rate offenders (relative to those without a removal) and the increasing risk of ICE detention and removal suppressed what might otherwise have been many more rearrests. It may also be that some unmeasured characteristics related to both repeat arrests and previous removal accounted for much or all of the relationships between them. More research is needed on this population using data that include immigration records of ICE apprehensions, detentions, releases, and removals during the parallel follow-up period. Absent access to this additional data source, only speculation is possible.

With the available data, we had a limited ability to further explore the present findings related to frequency of rearrest. In that effort, we compared the two groups on the number of days until last arrest (rather than first) in the 9-year follow-up period. Among the 369 individuals who were rearrested at least once, the last rearrest occurred significantly more quickly on average for those with a record of previous deportation. Specifically, over the 9-year (3,285 days) follow-up period, the average number of days until last rearrest among the previously removed was 1,118 days (or 3.6 years; SD = 899.6) compared with 1,627 days (or 4.5 years; SD = 967.9) for those never removed (t = 4.91, p < .001). The nearly 1 year difference in time-to-last rearrest could potentially reflect a greater relative risk of detection, detention, and removal for those with a record of prior removal. In other words, it is possible that a larger share of the previously removed become “ineligible” to recidivate later in the follow-up period by virtue of being identified and detained by ICE.

Despite its potential relevance to questions raised by Secure Communities, this study should not be interpreted as an assessment of the potential impact of the program. Secure Communities did not launch until 2008. By that point in the 9-year follow-up period, 70% of the individuals who were rearrested at least once had already experienced their final recorded rearrest. By the end of the (September) 2011 follow-up period, however, Secure Communities had been activated in approximately 50% of all U.S. jurisdictions (ICE, 2013c). Thus, this specific program may have had some unmeasured effect late in the follow-up period.

Another challenge related to study of this population is that “previous removal” is not a static underlying individual characteristic. Any unauthorized immigrant may acquire the designation of “previously removed” if, at any point, they are apprehended by ICE. Indeed, ICE data suggest this has become more likely as removals from criminal justice populations have steadily increased over the study’s follow-up period, from 81,626 in 2003 to 216,698 in 2011 (ICE, 2013b). This sizable increase suggests that a record of previous removal was a less common distinction in 2002 than it may be today. Consequently, it is likely that at least some individuals within the sample treated as members of the unauthorized immigrant comparison group acquired a record of removal (or were under final orders of removal) during the study’s follow-up period. Therefore, our findings must be interpreted as more relevant to conditions that produced a record of prior removal in or before 2002.

Given the limitations of conducting individual-level recidivism research with this population, it is reasonable to question its utility. Through the universal implementation of Secure Communities, willingly or not, local jurisdictions are now participating in locating unauthorized immigrants for potential detention or removal. Therefore, unless Secure Communities is substantially revisited by federal policy makers, the policy questions perhaps most relevant today (at least for state and local officials) are not whether local law enforcement should be involved in interior immigration enforcement; instead, they are (a) where their resources can be most effectively applied, and (b) the circumstances under which their active cooperation is supported by the communities they serve. These can be viewed as empirical questions that fall within the purview of criminologists to help address. To date, our field has conducted almost no work relevant to these questions, largely owing to the extreme difficulty in accessing individual-level data on officially determined immigration status that can be merged with criminal records. Indeed, the Los Angeles County Jail Foreign Born Inmates study (Hickman et al., 2005) is a rare example of this sort of data set.

While the present study alone by no means settles the question of whether previous removal is a consistent indicator of high criminal recidivism risk, it provides at least some empirical suggestion that this group may be worthy of local review and policy consideration. Los Angeles County is one jurisdiction that has conducted its own reviews of previously removed immigrant recidivism in its local context (CCJCC, 1990, 1997). As a result, it developed long-standing County-level policies around active cooperation with ICE, in the case of previously removed immigrants who are located within its jail population (CCJCC, 2002; Raymond et al., 2004). Much more research is needed before any definitive conclusions can be drawn about whether previous removal poses a distinct recidivism risk factor overall.

For now, state and local policy makers do not have the luxury of waiting (potentially decades) for researchers to build a conclusive body of evidence before deciding what to do about ICE requests for custody of specific arrestees through the Secure Communities process. The value of the present study is that it offers local and state policy makers some empirical evidence for consideration. For federal immigration authorities, the study offers an illustration of the potential value of sharing immigration records with academics for analysis and open publication. While the findings of the study offer support for ICE’s current enforcement priorities, support is limited because of the lack of follow-up immigration records. While ICE does occasionally provide data to researchers for analysis, the resulting studies are generally not peer-reviewed and may not be released publicly.

In building the body of research in this area, it is also necessary for other agencies within the federal government to take some responsibility for prioritizing and supporting immigration and crime research, as it has for research on many other crime-related topics posing pressing policy questions, such as community policing, domestic violence, and school violence.⁵ The latter crime-related topics fall primarily within the enforcement responsibilities of local criminal justice systems, rather than the federal government. Yet, federal support has helped create the partnerships and the funding for academics to actively engage with local criminal justice agencies to access their records, to study the performance and decision making of their personnel, and to collect data directly within their facilities and from individuals identified within their caseloads. It is hard to imagine the state of criminology today if we had never experienced this level of federal support for local criminal justice research. Given the federal role in immigration enforcement and control of the records critical to this area of study, a similar level of federal commitment is needed to advance the field in the direction of empirically informed policy guidance in the era of Secure Communities.

Footnotes

Acknowledgements

The authors gratefully acknowledge the assistance of the Los Angeles County Sheriff’s Department, the California Department of Justice Division of California Justice Information Services, and Dr. Greg Ridgeway.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biographies

Laura J. Hickman is a professor in the Division of Criminology & Criminal Justice of Portland State University and an Adjunct Behavioral Scientist with the RAND Corporation. The focus of Dr. Hickman’s research is on evaluating and improving criminal justice policy and practice. Her work covers a range of topics including capital punishment, violence amd victimization, corrections, and immigration.

Jennifer Wong is an assistant professor in the School of Criminology, Simon Fraser University. Her work employs applied and quantitative methods to study issues of delinquency/crime prevention and intervention, focusing primarily on crime prevention and crime control policy. Her recent work includes systematic reviews and meta-analyses in the areas of gang prevention programs and community-based prevention programs for youth.

Marika Suttorp-Booth is a statistical analyst in the Statistical Research and Consulting group at the RAND Corporation. Her current work includes both criminal justice and health research, including an evaluation of the economic cost of drug abuse and the Healthy Passages Longitudinal Study of Adolescent Health project. She is also a member of the Southern California Evidence-Based Practice Center.

References

Applegate

Chotiner

K. L.

(1999). Building LA’s Consolidated Criminal History Reporting System (CCHRS). National Center for State Courts, Education Session. Retrieved from http://cdm16501.contentdm.oclc.org/cdm/ref/collection/tech/id/649

Bang

Robins

J. M.

(2005). Doubly robust estimation in missing data and causal inference models. Biometrics, 61, 962-972.

Countywide Criminal Justice Coordination Committee. (1990). Criminal aliens in the Los Angeles County Jail population: Final report. Los Angeles, CA: Author.

Countywide Criminal Justice Coordination Committee. (1997). The impact of deportable aliens on the Los Angeles County Jail system: Final report. Los Angeles, CA: Author.

Countywide Criminal Justice Coordination Committee. (2002). High intensity criminal alien apprehension prosecution program: Fact sheet. Los Angeles, CA: Author.

Cullen

F. T.

Gendreau

(2000). Assessing correctional rehabilitation: Policy, practice, and prospects. In Horney

Martin

MacKenzie

D. L.

Peterson

Rosenbaum

(Eds.), Criminal justice 2000: Policies, processes, and decisions of the criminal justice system (Vol. 3., pp. 109-175). Washington, DC: Office of Justice Programs, U.S. Department of Justice.

Davis

R. C.

Erez

Avitabile

(2001). Access to justice for immigrants who are victimized: The perspectives of police and prosecutors. Criminal Justice Policy Review, 12, 183-196.

Dawson

(2011). How significant is a boxplot outlier? Journal of Statistics Education, 19(2), 1-13.

Decker

S. H.

Lewis

P. G.

Provine

D. M.

Varsanyi

M. W.

(2009). On the frontier of local law enforcement: Local policy and federal immigration law. In McDonald

W. F.

(Ed.), Sociology of crime, law and deviance, Vol. 13: Immigration, crime and justice (pp. 261-276). Bradford, UK: Emerald.

10.

Elliott

D. S.

(1995). Lies, damn lies, and arrest statistics. The Sutherland Award presentation to the American Society of Criminology meetings, Boston, MA. Boulder, CO: Center for the Study and Prevention of Violence, Institute of Behavioral Science, University of Colorado.

11.

Geerken

M. R.

(1994). Rap sheets in criminological research: Considerations and caveats. Journal of Quantitative Criminology, 10, 3-21.

12.

Harmon

(2012, March 17). Undocumented immigrants in jail: Who gets deported? Statesman Journal. Retrieved from http://www.statesman.com/news/statesman-investigates/undocumented-immigrants-in-jail-who-gets-deported-2244677.html

13.

Hickman

L. J.

Suttorp

M. J.

(2008). Are deportable aliens a unique threat to public safety? Comparing the recidivism of deportable and nondeportable aliens. Criminology & Public Policy, 7, 59-82.

14.

Hickman

L. J.

Suttorp

M. J.

(2010). The recidivism patterns of previously deported aliens released from a local jail: Are they high-risk offenders? Crime & Delinquency, 56, 507-534.

15.

Hickman

L. J.

Suttorp

M. J.

Wong

(2005). Deportable aliens released from the Los Angeles County Jail: A comparison of 1990, 1995, and 2002 release cohorts. Santa Monica, CA: RAND.

16.

Hosmer

D. W.

Lemeshow

(1999). Applied survival analysis: Regression modeling of time to event data. New York, NY: John Wiley.

17.

Huebner

(2009, April). Long-term recidivism patterns of different types of offenders. Paper presented at the Arizona State University Prisoner Reentry Seminar, Phoenix.

18.

Johnson

B. R.

(2004). Religious programs and recidivism among former inmates in prison fellowship programs: A long-term follow-up study. Justice Quarterly, 21, 329-354.

19.

Jordon

(2011, May 14). States rebel over deportations. The Wall Street Journal. Retrieved from http://online.wsj.com/article/SB10001424052748704681904576321404203166580.html

20.

Katz

C. M.

Fox

A. M.

White

M. D.

(2011). Assessing the relationship between immigration status and drug use. Justice Quarterly, 28, 541-575.

21.

Khashu

(2009). The role of local police: Striking a balance between immigration enforcement and civil liberties. Washington, DC: Police Foundation.

22.

Latessa

E. J.

Listwan

S. J.

Koetzle

(2013). What works (and doesn’t) in reducing recidivism. Waltham, MA: Anderson Publishing.

23.

Long

S. J.

(1997). Regression models for categorical and limited dependent variables: Advanced quantitative techniques in the social sciences. Thousand Oaks, CA: SAGE.

24.

MacKenzie

D. L.

(1997). Criminal justice and crime prevention. In Sherman

L. W.

Gottfredson

MacKenzie

D. L.

Eck

Reuter

Bushway

(Eds.), Preventing crime: What works, what doesn’t, what’s promising (pp. 9-1 -9-76). Washington, DC: National Institute of Justice.

25.

Maltz

M. D.

(1984). Recidivism. Orlando, FL: Academic Press.

26.

McCaffrey

Ridgeway

Morral

(2004). Propensity score estimation with boosted regression for evaluating causal effects in observational studies. Psychological Methods, 9, 403-425.

27.

McCullagh

Nelder

J. A.

(1989). Generalized linear models (2nd ed.). Cambridge, UK: Cambridge University Press.

28.

Morton

(2012, December 21). Civil immigration enforcement: Guidance on the use of detainers in the federal, state, local, and tribal criminal justice systems. U.S. Immigration and Customs Enforcement, Memorandum to All Field Office Directors. Retrieved from http://www.ice.gov/doclib/detention-reform/pdf/detainer-policy.pdf

29.

Office of Inspector General. (2010). The performance of 287(g) agreements (OIG-10-63). Washington, DC: U.S. Department of Homeland Security.

30.

Orr

J. M.

Sackett

P. R.

(1991). Outlier detection and treatment in I/O psychology: A survey of researcher beliefs and an empirical illustration. Personnel Psychology, 44, 473-486.

31.

Ostermann

(2013). Active supervision and its impact upon parolee recidivism rates. Crime & Delinquency, 59, 487-509.

32.

Ousey

G. C.

Kubrin

C. E.

(2009). Exploring the connection between immigration and violent crime rates in U.S. cities, 1980-2000. Social Problems, 56, 447-473.

33.

Passel

Cohn

(2012). Unauthorized immigrants: 11.1 million in 2011 (Pew Research: Hispanic Trends Project). Washington, DC: Pew Research Center. Retrieved from http://www.pewhispanic.org/2012/12/06/unauthorized-immigrants-11-1-million-in-2011/

34.

Petersilia

Turner

Fain

(2000). Profiling inmates in the Los Angeles County Jail: Risks, recidivism, and release options (RAND Report DRU-2394-NIJ). Santa Monica, CA: RAND.

35.

Powell

Perreira

K. M.

Harris

K. M.

(2010). Trajectories of delinquency from adolescence to adulthood. Youth & Society, 41, 475-502.

36.

Rao

J. N. K.

Scott

A. J.

(1981). The analysis of categorical data from complex sample surveys: Chi-squared tests for goodness of fit and independence in two-way tables. Journal of the American Statistical Association, 76, 221-230.

37.

Rao

J. N. K.

Scott

A. J.

(1984). On chi-squared tests for multiway contingency tables with cell proportions estimated from survey data. Annals of Statistics, 12, 46-60.

38.

Raymond

Hickman

L. J.

Williams

(2004). Identifying deportable aliens in the Los Angeles County Jail: Implementing the HI-CAAP federal-local partnership (WP-193-LAC). Santa Monica, CA: RAND.

39.

Ridgeway

(2006). Assessing the effect of race bias in post-traffic stop outcomes using propensity scores. Journal of Quantitative Criminology, 22, 1-29.

40.

Rosenbaum

J. L.

(1987). Social control, gender and delinquency: An analysis of drug, property and violent offenders. Justice Quarterly, 4, 117-132.

41.

Rosenbaum

Rubin

(1984). Reducing the bias in observational studies using subclassification on the propensity score. Journal of the American Statistical Association, 79, 516-524.

42.

Rubin

Blankstein

(2012, October 4). Chief Beck eases policy on illegal immigrant deportation. The Los Angeles Times. Retrieved http://articles.latimes.com/2012/oct/04/local/la-me-lapd-immigration-20121004

43.

Saunders

Lim

Prosnitz

(2010). Enforcing immigration law at the state and local level: A public policy dilemma (OP-273). Santa Monica, CA: RAND.

44.

Taxin

(2011, August 16). Immigrants plead for end to fingerprint sharing. Houston Chronicle. Retrieved from http://www.huffingtonpost.com/2011/08/16/immigrants-plead-for-end-_n_927965.html

45.

U.S. Government Accountability Office. (2011). Criminal alien statistics: Information on incarceration, arrests, and costs (Report to Congressional Requesters). Washington, DC: Author. Retrieved from http://www.gao.gov/new.items/d11187.pdf

46.

U.S. Government Accountability Office. (2012). Secure communities: Criminal alien removals increased, but technology planning improvements needed (Report to the Ranking Member, Committee on Homeland Security, House of Representatives). Washington, DC: Author. Retrieved from http://www.gao.gov/assets/600/592415.pdf

47.

U.S. Immigration and Customs Enforcement. (2012). Activated jurisdictions: Secure Communities. Washington, DC: U.S. Immigration and Customs Enforcement, U.S. Department of Homeland Security. Retrieved from http://www.ice.gov/doclib/secure-communities/pdf/sc-activated2.pdf

48.

U.S. Immigration and Customs Enforcement. (2013a). Activated jurisdictions. Washington, DC: U.S. Immigration and Customs Enforcement, U.S. Department of Homeland Security. Retrieved from http://www.ice.gov/doclib/secure-communities/pdf/sc-activated.pdf

49.

U.S. Immigration and Customs Enforcement. (2013b). Removal statistics. Washington, DC: U.S. Immigration and Customs Enforcement, U.S. Department of Homeland Security. Retrieved from http://www.ice.gov/removal-statistics/

50.

U.S. Immigration and Customs Enforcement. (2013c). Secure Communities: Monthly statistics through May 31, 2013. Washington, DC: U.S. Immigration and Customs Enforcement, U.S. Department of Homeland Security. Retrieved from http://www.ice.gov/doclib/foia/sc-stats/nationwide_interop_stats-fy2013-to-date.pdf

51.

Vogel

McDonald

W. F.

Jordan

Düvell

Kovacheva

Vollmer

(2009). Police cooperation in internal enforcement of immigration control: Learning from international comparison. In McDonald

W. F.

(Ed.), Sociology of crime, law and deviance, Vol. 13: Immigration, crime and justice (pp. 205-241). Bradford, UK: Emerald.

52.

Wadsworth

(2010). Is immigration responsible for the crime drop? An assessment of the influence of immigration on changes in violent crime between 1990 and 2000. Social Science Quarterly, 91, 531-553.

53.

Warner

J. A.

(2005). The social construction of the criminal alien in immigration law, enforcement practice and statistical enumeration: Consequences for immigrant stereotyping. Journal of Social and Ecological Boundaries, 1(2), 56-80.

54.

Wong

J. S.

Schonlau

(2013). Does bully victimization predict future delinquency? A propensity score matching approach. Criminal Justice and Behavior, 40, 1184-1208.

55.

Worrall

(2008). Crime control in America: What works? (2nd ed.). New York, NY: Pearson.