Administrative Discretion Through Changing Presidencies and Political Polarization: Reflection on the Rise, Fall, and Rise of Federal-Local Immigration Partnerships in the U.S.

ABSTRACT

Created by Congress in 1996, the immigration enforcement program 287(g) has experienced several cycles of rapid expansion and contraction during different stages in the policy’s relatively brief lifetime. It is an intergovernmental program where state and local agencies sign onto agreements with U.S. Immigration and Customs Enforcement (ICE), effectively deputizing law enforcement agents to implement federal immigration policy. Using logit models and the lens of bureaucratic discretion, we find that agencies in politically conservative jurisdictions and counties with proportionally large Hispanic populations were most likely to sign these agreements and maintain them during the program’s lean years. Conservative counties were even more likely to have them when they have larger Hispanic populations and during times of Republican presidential administrations. This paper puts these findings in comparative perspective and links them to the worldwide trend of democratic decline and the rise of right-wing populism in Europe and elsewhere.

KEYWORDS:

• Administrative discretion
• policy implementation
• immigration policy
• 287(g)
• cooperative federalism

Introduction

When it comes to U.S. policy, perhaps no immigration initiative has been more controversial than the 287(g) program (Brown, 2021; Hagan et al., 2011, p. 1,376). Named after Section 287(g) of the Illegal Immigration Reform and Immigrant Responsibility Act (IIRIRA) of 1996, the program has received a great deal of scrutiny partly because it allows state and local law enforcement to perform certain immigration policing functions normally performed by federal officials (e.g., Capps et al., 2011; Creek & Yoder, 2012; Wong, 2012). Immigration and Custom Enforcement (ICE) has described the program to Congress and the public as a means of apprehending and deporting dangerous, high-level criminal aliens (U.S. Department of Homeland Security [DHS], 2011). Yet, various minority rights and civil liberty groups have complained that the enforcement of 287(g) has resulted in the abuse of rights for legal and undocumented immigrants alike (Capps et al., 2011).

The 287(g) program is a cooperative-federalism immigration program that permits local law enforcement agencies, primarily county sheriff offices and city police, to enter into memorandums of agreement (MOA) with ICE to apprehend and detain individuals suspected of being in the U.S. without proper documentation. Here, we examine local decisions to adopt 287(g) in the United States over the years 2006 through 2020. The program experienced initial growth after the terrorist attacks of September 11 2001, contraction during the Obama Administration to a mere 32 MOAs nationally, and then a rapid expansion under the Trump Administration to 148 MOAs.

We explore the program’s evolution through the lens of administrative discretion and how the use of this discretion by local-level administrators interacts with national-level politics. We do this by examining the conditions in which local administrators chose to enter into 287(g) agreements with the federal government. We posit that the program’s expansion can be explained, at least in part, as an interaction between national politics and local conditions. Local administrators felt compelled to participate in the federal program in politically conservative communities where anti-immigration sentiments likely run high and did so more often during times when conditions at the national level facilitated such participation. Our thesis is confirmed with time-series panel logit models, which demonstrate that law enforcement agencies in conservative communities were significantly more likely to enter 278(g) MOAs, as were agencies with large Hispanic populations. Additionally, communities were more likely to adopt the program during Republican presidential administrations. Moreover, these factors compound each other. More conservative counties with larger Hispanic populations were more likely to have active agreements than were conservative counties with smaller Hispanic populations. Conservative counties were also more likely to have them when a Republican was in the White House than when a Democrat was president.

The implications of the findings extend in two important ways: (1) they extend geographically, as the discretionary practices of various civil servants that go beyond the local level will be better understood when the idiosyncrasies of local practices are unraveled and the interactions amongst various governance levels are better understood nationally and internationally (Oyekanmi & Agboola, 2022), and (2) they will have implications to various social scientific fields as researchers might want to reassess the consequences emanating from recent trends of ethno-national movements and the seeming legitimization of behavioral patterns at various governance levels as migrants worldwide have come to be “perceived as a threat to the social order” (Ambrosini et al., 2020, p. 21).

LEAs and the evolution of 287(g)

Scholars note there has been a recent trend of devolution, or “localizing,” immigration policy in both the United States and Europe (Yeo & Huang, 2020). This idea of state and local law enforcement officers enforcing aspects of federal immigration policy is relatively new in the United States. After the Civil War, the power to set and enforce rules governing who can be in the country has rested exclusively with the federal government.¹ States could, and frequently did, pass overtly racist policies that limited the rights of immigrants and individuals with non-European ethnicities, such as California’s Alien Land Law (1913), which prohibited Asians from owning or possessing long-term leases on land (Ferguson, 1947). However, issues of one’s immigration status, such as whether a person was in the country legally or could become a U.S. citizen, were matters of federal statute. Further, one’s immigration status was, and remains, a civil matter, not criminal.² Thus local, law enforcement agencies (LEAs) had no role in implementing federal immigration policy for most of the county’s history. This situation began to change with passage of the Illegal Immigration Reform and Immigrant Responsibility Act (IIRIRA) of 1996. Section 287(g) of IIRIRA authorized state and local LEAs to sign Memorandums of Agreement (MOAs) with the federal government to perform certain federal immigration functions (ICE, 2006).

Although 287(g)’s life technically began in the 1990s, it was not until after the terrorist attacks of September 11 2001 that LEAs began utilizing the program. Initially, the official position of the U.S. Department of Justice (DOJ) was that state and local LEAs only had the authority to enforce criminal violations of immigration laws (Pham, 2018). This changed in June 2002 when George W. Bush’s Attorney General John Ashcroft announced the DOJ was asking state and local LEAs to enforce “civil provisions [of immigration law] that render an alien deportable” (Ashcroft, 2002). In other words, the Bush DOJ was requesting help from non-federal law enforcement officers to find and detain undocumented immigrants. However, the DOJ, at least initially, only sought to expand the power of state and local LEAs to enforce civil immigration policy narrowly. As Ashcroft (2002) explained, “[t]he Department of Justice has no plans to seek additional support from state and local law enforcement in enforcing our nation’s immigration laws, beyond our narrow anti-terrorism mission.” Fittingly, the first 287(g) MOAs were signed with statewide agencies and narrowly focused on combatting terrorism beginning in 2002. However, the program did not remain narrowly focused. By the end of the Bush Administration there were 59 MOAs with non-federal LEAs, and most individuals detained through the program had only minor criminal infractions (e.g., driving without a license) or no criminal violations whatsoever (Capps et al., 2011, pp. 52–53).

These early agreements followed one of three models: the jail model, the task force model, and the hybrid model. Under the jail model, police could only inquire into an individual’s immigration status within the confines of the jail, after the person had been arrested or, in some cases, convicted of a crime (Capps et al., 2011). The task force model placed greater authority in trained officers, empowering them to inquire into an individual’s immigration status and issue detainers in the field. They could also issue arrest warrants and execute search warrants. The hybrid model, which was rare, combined both approaches (Capps et al., 2011).

In the first years of the Obama Administration, the program continued expanding, with 67 MOAs by the end of 2009, Obama’s first year in office (Chacon, 2009, p. 1582). From fiscal year (FY) 2006 through 2010, 287(g)-trained LEA officers screened more than 186,000 immigrants for potential removal. By the end of the 2010 FY, the program accounted for about 10% of individuals identified for removal by DHS (Capps et al., 2011).

As the program grew, so did scrutiny of it. The U.S. Government Accountability Office (GAO) documented serious lack of oversight and training for the program (GAO, 2009). Civil liberties and immigrant rights groups argued that 287(g) encouraged racial profiling (American Civil Liberties Union [ACLU] 2012; Lacayo, 2010). Immigration attorneys, and even immigrants, themselves, stated that the program created a fear of working with law enforcement among immigrants and immigrant family members. They argued expanding the role of traditional police to cover civil immigration violations will make undocumented immigrants, or their loved ones, afraid to report crimes – even when immigrants, themselves, are the victims (Chandler, 2008; Weissman et al., 2009). Others criticized the program’s vague policy objectives. Participating law enforcement exercised varying levels of application of 287(g). Some focused on serious criminal immigrants, but others used the program to catch as many unauthorized immigrants as possible, regardless of their criminal records (Capps et al., 2011). Some LEAs that participated in the program explicitly stated they would only target “illegal immigrants who commit crimes,” only to later use the expansive authority to target all undocumented immigrants (Weissman et al., 2009, pp. 27–28).

Responding to complaints, the Obama Administration halted 287(g)’s expansion in 2009 to focus on reforming the program. Changes to the program were ostensibly designed to focus on the identification and removal of serious criminals and other public safety threats (Capps et al., 2011). In December 2012, Obama’s ICE Director John Morton issued a memo announcing that it would not renew agreements to use the Task Force approach in 2013 (ICE, 2012). Also in 2013, ICE’s much larger Secure Communities (S-Comm) program, which screens the biometric information of individuals booked into local jails, went operational nationwide. With the intense criticism of 287(g) and ICE’s shifting priorities toward S-Comm, the size of the 287(g)-program declined significantly. At the end of Obama’s second term, there were a mere 32 agreements nationwide (Pham, 2018, p. 1274).

Yet, Donald Trump, who ran on decidedly anti-immigration agenda in 2016, breathed new life into the program. Less than one week into his administration, Trump signed Executive Order (EO) 13767 ordering DHS to enter into more 287(g) MOAs and to structure such agreements “in a manner that provides the most effective model for enforcing Federal immigration laws…”³ Shortly thereafter, then-DHS Secretary John Kelly issued a memo directing the Director of ICE and the Commissioner of Customs and Border Protection (CBP) to “expand the 287(g) program… to the greatest extent practicable” (Kelly, 2017, p. 4). For the first time, LEAs were allowed to sign MOAs with ICE and CBP to enforce immigration policy.⁴

In the months that followed, the program grew rapidly in size. In Trump’s first year in office, 25 new LEAs signed agreements to participate, 18 of which were in Texas (ICE, 2017; Pham, 2018, pp. 1,274). According to a recent ACLU study, 59% of sheriff offices to sign MOAs during this period had records of “anti-immigrant rhetoric” and nearly two-thirds had records of racial profiling or other civil liberties abuses (ACLU, 2022). Trump’s DHS also slightly restructured the program. Initially, Trump expressed interest in reviving the old Task Force Model; however, the administration ultimately settled on keeping the Jail Model and developing a new, and highly similar, Warrant Service Officer (WSO) Model. Under WSO agreements, officers are permitted to perform most of the same functions as the Jail Model, such as arresting undocumented immigrants, even when they face no criminal charges. Yet, officers in WSO agencies are not permitted to interrogate an individual about their immigration status if the officer believes the individual is undocumented (American Immigration Council [AIC], 2021). By the time Trump left office in January 2021, ICE announced it had 148 MOAs in place: 72 Jail Models in 21 state and 76 WSO Models in 11 states (ICE, 2021). As of June 2021, only one of the new MOAs had been terminated (AIC, 2021). The Trump Administration’s aggressive interior immigration enforcement posture faced some public backlash. Three years into his presidency, polls showed 60% of Americans opposed removing all undocumented immigrants, with 80% supporting opportunities for them to become U.S. citizens (Norman, 2019). Additionally, some local governments that adopted 287(g) MOAs later suspended the agreements after immigrants and immigrants rights groups voiced concerns (Lyle, 2019).

Understanding factors behind 287(g)’s expansion, contraction, and subsequent re-expansion contributes to our broader understanding of our modern “multilayered jurisdictional patchwork of immigration federalism” (Varsanyi et al. 2021). Unlike most cooperative federalism programs, such as S-Comm, 287(g) does not merely request non-federal actors assist federal authorities by exercising their pre-existing authority (Pham, 2018, p. 1258). Rather, LEAs with 287(g) agreements are granted new powers that in all other cases are reserved for federal officials. Most notable of these is the power to arrest and detain immigrants for nothing more than civil immigration violations.

Obviously, changes in administration have affected the implementation of 287(g). Yet, changes in the presidency cannot alone account for the program’s roller-coaster implementation. In the following section, we discuss the theory of administrative discretion, the lens through which we explore the evolution of 287(g) and explore the exercise of discretion in immigration policy through a comparative context.

Administrative discretion in a comparative context

When compared to topics such as health care, environmental management, or family policy, immigration is a relatively understudied topic within public administration (Yeo & Huang, 2020). We build upon the growing scholarly body of research examining the exercise of administrative discretion in immigration policy (e.g., Chand & Schreckhise, 2015; Edlins & Larrison, 2020; Larrison & Edlins, 2020; Pedroza, 2019). Writing about street-level bureaucrats, Lipsky (1980), depicted these individuals as “in many ways the most powerful actors in the policy [implementation] process” (Chand & Schreckhise, 2015, pp. 1, 622). This discretionary power is also viewed as a dilemma, with bureaucrats inescapably forced to exercise discretion over major policy decisions (Hudson, 1993). In areas such as health care or social services, administrators often use discretion to fulfill an agency’s clientele needs within a policy’s guidelines (Tummers & Bekkers, 2014). However, law enforcement is inherently different from social service policy. Law enforcement officers in difficult situations frequently weigh the law against personal values to exercise discretion in situations where moral judgement is concerned (Maynard-Moody & Musheno, 2022).

In the case of immigration policy, bureaucrats with administrative discretion, to a significant extent, rely on their personal values and experiences. In addition, “organizational socialization reproduces and sustains institutionalized social interactions between the bureaucrat and their clients” (Borrelli, 2021, p. 579). When surveying public administrators’ attitudes in Arizona localities towards noncitizens, both documented and nondocumented, Lucio (2016) finds that public safety administrators were less inclined to be supportive of undocumented immigrants’ civic participation. Policy areas that are deemed controversial and decision-making processes that are characterized by high complexity, often give rise to high levels of discretionary power being granted to civil servants and administrators in the field. The individual characteristics of their target populations can influence the decisions these administrators make, including the population’s race and ethnicity (Chand, 2020; Meier, 1993; Watkins-Hayes, 2011; Wilkins & Williams, 2008). Studying the role of these street-level bureaucrats in immigration policy becomes essential, as their discretionary power in the decision-making process is highly consequential to the functioning of social microcosms (Bouchard and Carroll, 2002). As Edlins and Larrison (2020) explain in their study of street-level bureaucrats’ interactions with a uniquely positioned immigrant group, unaccompanied minors, this specific group is “fully dependent on … [street-level bureaucrats] for care” (p. 420).

Putting administrative discretion in comparative perspective, the landscape of administrative discretion might be changing based on push and pull factors and what societal actors perceive as legitimate or mainstream in the local political space. Bearing in mind that the world has been living through sixteen years of consecutive democratic backsliding, populism, and political extremism having gained momentum, it is perhaps not surprising that migrants have come to be “perceived as a threat to the social order,” which, in turn, gives rise to “far right movements find[ing] legitimization and expand[ing] their audience mobilizing together with local residents and authorities against” them (Ambrosini et al., 2020, p. 21).

As Ambrosini and colleagues argue, street-level bureaucrats who practice administrative discretion have experienced pressure from civil societal actors in their local communities from opposing directions: 1) political mobilization opposing immigration and asylum seekers associated with far-right parties, and 2) mobilizing “advocacy coalition[s] in favor of refugees and migrants, ranging from radical social movements to trade unions, to religious institutions, to common citizens and spontaneous groups” (2020, p. 21). In this “battleground” of “local policies of exclusion,” where “the opposition against refugees recreates social bonds,” local bureaucrats may find themselves in a proverbial pickle (Ambrosini et al., 2020, pp. 20–21). Who the street-level bureaucrats find themselves siding with will largely be determined by their own political views and affiliations. However, what they will act on and how they will use their discretion will be likely conditioned by the prevalent local political culture and the current political affiliation of the locality. In the context of U.S. immigration federalism, where state and local governments wield significant influence over policy, Chand and Schreckhise find, “local political attitudes play a role, with Republican-leaning jurisdictions and those in states that support restrictive state-level immigration witnessing more deportations” (2015, p. 1621). Still, it is unlikely that political conservativism alone would explain local adoption of 287(g). Despite most U.S. counties voting Republican, there is evidence that anti-immigration localities, those supporting punitive unwelcoming policies, tend to be the outliers (Williamson, 2018). Indeed, a somewhat recent survey data of police chiefs finds that most support policies that are welcoming or, at least, neutral to immigrant integration (Williams, 2015).

In different places this plays out in various ways, depending on the prevalent local political climate. Examining Spanish immigration policy, Bastien echoes Lipsky’s theory as she confirms that street-level bureaucrats shape public policies through exercising their discretion in implementation (2009, p. 665). Bastien observes that there is a “link between goal ambiguity and the use of informal discretion,” albeit with variances depending on whether such ambiguities were prevalent more on the organizational or at the individual level (2009, pp. 665, 680). Contextualizing immigration policy implementation in a handful of European nations, Jordan et al. (2003) confirmed that, while organizational culture matters, the overall complexities of immigration policies and policy implementation are enormous since political culture, existing and emerging ethnic identities all seem to have important implications on how discretion is exercised. Policy implementation in immigration matters for numerous reasons, but even “potential economic benefits of migration are filtered through the interpretations and practices of staff in these very varied agencies, with their particular cultures and organizational perceptions of both national interests and immigrant identities” (Jordan et al., 2003, p. 206). While the European Union has proven to demonstrate characteristics of a supranational entity in some regard, as Martinsen argues, immigration policy and “entry and residence rights of third-country nationals” within the European Union, are filtered “through a wide web of conditions and administrative discretion to such an extent that the policy domain has long been one in which national sovereignty seems to remain intact” (2011, p. 953).

Thus, perhaps it should come as no surprise that Trump’s running on an ethno-national agenda has had a perhaps undue influence on the Republican Party’s immigration policy implementation practices, as sympathizing street-level bureaucrats, consciously or subconsciously, might have found that their deeply held personal values were being validated. In the past, they might have not acted upon those and still be at ease, but it might have changed under new circumstances. As Ambrosini states, these civil servants and other bureaucrats have “power to interpret the rules and procedures when it comes to applying them to specific cases” and characterizes them as “discretional gate-keepers” (Ambrosini, 2017, p. 13).

As far-right populism increases in other western democracies, examining local implementation in the United States helps to develop theory that can broadly explain the exercise of administrative discretion within the context of growing anti-immigration sentiment.

Data and hypotheses

To examine how these subnational and national forces interact regarding immigration enforcement in the U.S., we examine what can account for the variation across time and location of counties signing agreements with the federal government to assist with its immigration enforcement efforts. We employ as our dependent variable whether the county for each year has an active signed memorandum of agreement with Immigration and Customs Enforcement for years 2006 through 2020. The unit of analysis is the county-year. Although the 287(g) program officially began in 1996 with the passage of IIRIRA, the program was unfunded until after 9/11 and ICE began signing MOAs with local law enforcement agencies. The first six counties, as opposed to the statewide agencies mentioned earlier, signed these in 2006. As Figure 1 displays, the trend in the number of counties who signed MOAs continued to rise until 2009, staying mostly level for the next two years, dropping between 2011 and 2016, then increasing again after 2016, reaching a total of 139 counties in 2020.⁵

Figure 1. Number of counties with signed 287(g) memoranda of agreement by year, 2006–2020.

Source: U.S. Immigration and Customs Enforcement, 2006–2020.

We begin by examining the relationship that the local political environment and the size and growth of their Hispanic populations have with the propensity for local county sheriff offices to sign these MOAs with ICE. Our models include a county-level Conservatism variable, adapted from Chand and Schreckhise (2015). For years when there was a presidential election, this value indicates the percentage of votes cast in the county that the Republican presidential candidate received. In years when there was no presidential election, the value for each county-year is an average between what the Republican candidate received in the most recent election what the candidate received in the successive presidential election.

Our models also include the variable %Hispanic which reflects the percentage of each county’s population that is Hispanic. These estimates that were collected from the Population Division of the U.S. Census Bureau for each county and year for the years 2006 through 2020. To determine the value for each county-year’s Hispanic population change variable (%Hispanic Δ), we also computed the percentage-point increase of the county’s Hispanic populations’ growth as a function of the county’s population compared to the Hispanic percentage of the county’s population county at time t compared to ten years earlier at t-10 (see Pedroza, 2019).

With these data, we hypothesize the following:

H1:

Conservative counties will more likely sign 287(g) MOAs with ICE.

H2:

Counties with larger Hispanic populations will more likely sign 287(g) MOAs with ICE.

H3:

Counties with growing Hispanic populations will more likely sign 287(g) MOAs with ICE.

H4:

Counties will be more likely to have active 287(g) MOAs with ICE during a Republican presidential administration than during Democratic presidential administration.

We include in our models additional social and economic variables that have been linked with the propensity for local law enforcement agencies to sign 287(g) MOAs with ICE or to cooperate with ICE in other ways. Wong (2012) found that larger counties and ones that experienced more recent population growth were more likely to sign the 287(g) MOAs. We include in our models the counties’ overall population and the variable Growth, which is the proportional growth of the counties that specifically reflects the percentage point increase in the size of population of each county from 10 years earlier. Wong (2012) also found that counties with larger populations of African Americans were less likely to sign 287(g) MOAs, as were counties with less unemployment. Counties with larger populations and those located in southern states were more likely to sign them.⁶ Schreckhise and Chand (2021) found counties with larger local law enforcement agency (LEA) budgets (and with the greater capacity presumedly larger budgets bring) removed more individuals under ICE’s Secure Communities program. We similarly include these data as per capita amounts reported by county law enforcement organizations for 2008, the last year nationwide county-level data are available. Schreckhise and Chand (2021) also found some evidence that counties with higher median per capita incomes and those located in along the U.S.-Mexico border removed more individuals under that program, as well.⁷ Finally, to determine whether counties are more likely to have signed MOAs during Republican presidential administrations, we include the variable Republican pres. See the appendix for the variables’ summary statistics.

Because our data are in panel format for each U.S. county across years 2006–2020 and include a binary dependent variable, we employ panel logit models. We use these instead of survival models because roughly 14% of the counties which signed MOAs during the time we examined either opted out of their MOAs or the counties failed to sign a new MOA after their previous MOA expired.⁸

Analysis

Table 1 presents two models predicting whether each county-year had an active 287(g) MOA. As expected, each of the models in Table 1 reveal that more conservative counties were more likely to have signed 287(g) memoranda of agreement with ICE, providing evidence for Hypotheses 1. This is illustrated in Figure 2, which displays the probablity having an MOA as count conservatism increases. It should be noted that Table 1 presents two separate models, examining separately the variables %Hispanic and %Hispanic Δ because these variables are highly correlated. Model 1 shows that counties with larger Hispanic populations were more likely to have signed MOAs, providing evidence for Hypotheses 2. The same is not true for counties with growing Hispanic populations, providing contrary evidence for Hypotheses 3 (see Model 2). Both models in this table also reveal that the counties more likely to sign MOAs had larger overall populations, and more Black residents. This is also true for counties located in the south, counties with larger law enforcement budgets, and higher median incomes. Additionally, counties were more likely to have active MOAs during Republican presidential administrations, providing clear evidence for Hypothesis 4.

Figure 2. The relationship between conservatism and the probability of an active 287(g) MOA.

Shaded area reflects 95% confidence intervals. Estimates are obtained from Model 1 in Table 1.

Table 1. Logit models of 287(g) counties.

	Model 1	Model 2
Conservatism	28.668***	26.060***
	(2.678)	(2.397)
%Hispanic	14.383***
	(1.951)
%Hispanic Δ		3.860
		(4.237)
Population (ln)	3.080***	3.133***
	(0.272)	(0.261)
Growth	0.306	−0.737
	(0.928)	(1.006)
%Black	13.200***	7.333*
	(3.186)	(2.958)
Southern	3.045***	4.274***
	(0.720)	(0.690)
Border county	−1.662	5.262*
	(1.801)	(2.334)
LEA budget	0.004**	0.005***
	(0.001)	(0.001)
Median income	1.398***	1.531***
	(0.125)	(0.127)
Unemployment	0.390***	0.392***
	(0.040)	(0.039)
Republican pres.	1.208***	1.214***
	(0.177)	(0.177)
Intercept	−83.083***	−79.984***
	(5.380)	(4.942)
No. of observations	46,257	46,257
No. of counties	3,085	3,085
No. of years	15	15
LL	−1,378.059	−1,403.055
df	13	13
AIC	2,782.118	2,832.111
BIC	2,895.764	2,945.757

Standard errors appear in parentheses.

***p < .001; **p < .01; *p < .05.

Figure 1 reveals a plot of the predicted probability of a county having an active 287(g) MOA for the years examined, using estimates from Model 1. It shows that very few counties with Conservatism scores of less than 0.4 have active 287(g) MOAs. However, as one moves to the right-hand side of the x-axis, the probability increases in a rather dramatic fashion, especially in counties with Conservatism scores greater than 0.7.

We suspect that the conservatism of a county and size of its Hispanic population could have a nonlinear relationship with the likelihood the county has a signed 287(g) MOA. Specifically, we suspect that the influence of Conservatism will be greater in counties with larger Hispanic populations. This is because it seems likely conservative counties with a large resident Hispanic population would more likely have an active MOA than would be the case in liberal counties with similarly large Hispanic populations. To test this possibility, we include in Model 3 (Table 2) the variables originally included in Model 1, but also include the interaction term “Conservatism x %Hispanic.” Our model reveals this is the case with a positive coefficient for the interaction term. More conservative counties become even more likely to sign 287(g) MOAs when they have larger Hispanic populations.⁹

Table 2. Logit models of 287(g) counties with interaction terms.

	Model 3	Model 4
Conservatism	20.498***	22.546***
	(2.967)	(2.825)
%Hispanic	−15.308*	14.580***
	(6.975)	(2.038)
Population (ln)	3.380***	3.125***
	(0.277)	(0.284)
Growth	−0.073	0.844
	(0.980)	(0.948)
%Black	11.702***	13.984***
	(3.194)	(3.279)
Southern	3.306***	3.133***
	(0.760)	(0.736)
Border county	1.524	−1.423
	(2.350)	(1.823)
LEA budget	0.004**	0.004
	(0.001)	(0.002)
Median income	1.427***	1.494***
	(0.131)	(0.131)
Unemployment	0.387***	0.394***
	(0.040)	(0.040)
Republican pres.	1.173***	−3.980***
	(0.181)	(0.701)
Conservatism x %Hispanic	47.887***
	(10.266)
Conservatism x Republican pres.		8.945***
		(1.191)
Intercept	−81.723***	−81.938***
	(4.980)	(5.576)
No. of observations	46,257	46,257
No. of counties	3,885	3,885
No. of years	15	15
LL	−1367.5121	−1346.484
df	14	14
AIC	2763.024	2720.968
BIC	2885.412	2843.355

Standard errors appear in parentheses.

***p < .001; **p < .01; *p < .05.

We display this interaction term graphically in Figure 3. The three lines reflect the margins for counties who have (1) zero Hispanic individuals living in their borders; (2) a percentage of the population at the mean for the state-years; and (3) those county-years with Hispanic populations, one standard deviation unit above the mean. The figure reveals that for all counties with Conservatism scores below 0.5, the probability of having a 287(g) MOA is quite small. However, as one moves towards the right-hand side of the x-axis, the probability increases, yet that probability does not increase equally across the three groups. Notably, the counties with Hispanic populations one standard deviation above the mean see a far greater likelihood of having an active 287(g) MOA than those at the mean and those counties that report no Hispanics. Hence, conservative counties with larger Hispanic populations are indeed more likely to have an active MOA than conservative counties will smaller Hispanic populations.

Figure 3. The relationship between conservatism, hispanic population, and the probability of an active 287(g) MOA.

Shaded area reflects 95% confidence intervals. Estimates are obtained from Model 3 in Table 2.

To what extent is there a link between the presence of a Republican in the White House, county conservatism, and the propensity for a county to have a signed MOA? As Figure 1 suggests, the number of counties with signed 287(g) MOAs grew modestly during the George W. Bush administration. The number dropped during the Obama administration, then grew relatively significantly during the Trump administration. The coefficients indicating a Republican presidential administration in models 1 and 2 clearly reveal such a link exists. We examine the relationship this growth had with a county’s ideological orientation to determine whether this growth was more likely to occur in conservative counties. Model 4 includes the interaction term “Conservatism x Republican pres.” It is significant and positive, indicating that when a Republican president is in office, conservative counties were more likely to have an active MOA. Figure 4 presents this relationship graphically. There is very little difference in the probability of a county having an active MOA in counties with Conservatism scores of less than 0.5. However, when a Republican is in the White House, conservative counties (i.e., those with Conservatism scores greater than 0.5) are more likely to have an active 287(g) MOA than they are when a Democrat is in the White House.

Figure 4. The relationship between conservatism, presidential party, and the probability of an active 287(g) MOA.

Shaded area reflects 95% confidence intervals. Estimates are obtained from Model 4 in Table 2.

Because of the panel nature of our data, we consider the possibility that the time-invariant variables may have an outsized influence on the results presented in models 1 through 4. To test this possibility, a series of models of counties adopting 287(g) MOAs were ran that included various combinations of the three time-invariant variables included in models 1 through 4: Southern, Border County, and LEA Budget. We do this to determine whether the exclusion of each of these variables, either individually or in combination with the other time-invariant variables, influences the results of each of our variables of interest in each of models 1 through 4. The results of these models are included in Appendix C. We find the exclusion of these variables does not alter the results presented in our models. Thus, it can be said with a degree of confidence that the time-invariant variables are not influencing the results presented above.

We also consider the role the trending behavior of time-variant variables may have on our results, notably, the possibility that the effect of some covariates have on our dependent variables may be heterogenous across time, resulting in the impact of our variables of interest appearing to have either greater (or possibly weaker) predictive power for the full 15-year time period examined than they, in fact, do have. To test for this possibility, we employ a series of models, one for each of our models 1 through 4 above, with year fixed effects. These are included in Appendix D. Although we find that the coefficient in Model 2 for the variable %Hispanic is no longer reaches conventional levels of statistical significance in the year fixed effects models, our remaining variables of interest do reach these levels, including the year fixed effect for Model 3 examining the Conservatism-%Hispanic interaction term. Thus, we can conclude that regardless of the timeframe examined within the 15-year span of our data, conservative counties with larger Hispanic populations are more likely to adopt 287(g) MOUs than liberal counties with relatively large Hispanic populations. Along similar lines, we include a trend term in our models in Appendix E to account for any potential trending behavior of our variables. We also run county fixed effects models in Appendix F to ensure our variables’ influences are evenly distributed across space.¹⁰ Again, we do not find any notable changes to our independent variables of interest in either set of results.

Discussion and conclusion

Last year marked the 20^th anniversary of 287(g)’s initial implementation following the 9/11 attacks. In its two-decade existence, it has come to be known as ICE’s “most controversial” program (Hagan et al., 2011, p. 1376), with its local implementation characterized through cycles of expansion and contraction. While this pattern is somewhat explained by executive-level changes nationally, there are also local explanations for the program’s roller-coaster implementation. We find that local political values and ethnic demographic makeup are prime explanations for the program’s adoption. LEAs in ideologically conservative communities or communities with large Hispanic or black populations are significantly more likely to opt into the program. An agency’s budget also predicts whether the county has MOAs, suggesting that LEAs with more resources are more likely to cooperate with ICE, which some prior research indicates (Jaeger, 2016).

The fact that conservative LEAs, and ICE itself, take more aggressive immigration enforcement actions in more political conservative communities has been established in prior research (Chand & Schreckhise, 2015; Schreckhise & Chand, 2021; Wong, 2012). Inversely, recent research has demonstrated that LEAs in conservative communities are significantly less likely to adopt pro-immigration “sanctuary policies” (D. Chand et al., 2022). It has also long been argued that the adoption or implementation of restrictive immigration policies can be explained by the presence of a large, or growing, nonwhite population (Rocha & Rodolfo, 2009). Yet, here we find that political ideology and ethnic makeup interact in ways unexplored by the prior academic literature on local immigration enforcement. The effect of local ethnic makeup is mitigated by political ideology. Specifically, the effect of Hispanic population is most strongly a predictor of 287(g) adoption in conservative communities (see Model 4). Additionally, the effect of a Republican presidency is also most pronounced in conservative communities. While 287(g) expanded under Republican administrations, LEAs in conservative counties were significantly more likely to adopt 287(g) under Republican administrations, as shown in Model 4.

The choice to enter 287(g) MOAs rests with the management of local law enforcement agencies. Scholars have sought to examine how such administrators exercise administrative discretion and policy making decisions (Lipsky, 1980), including some who have specifically examined discretionary decision making in an immigration policy context (e.g., Bouchard & Carroll, 2002; Chand & Schreckhise, 2015). Based on the findings here, it appears that local administrators are responding to local political attitudes and values when choosing to opt into intergovernmental immigration programs such as 287(g). Law enforcement agencies are significantly more likely to sign into 287(g) MOAs in conservative areas with strong anti-immigration sentiment. We believe this is not exclusively a U.S. phenomenon. The recent decade has witnessed a growth of right-wing populist movements in the West. In Europe, there is evidence that immigration administrators have opted to exercise administrative discretion in a more punitive manner in response to local populist movements (Ambrosini, 2020).

Public attitudes related to immigration ebbed and flowed across time, often in response to other issues such as the economy and national security, which typically have little, if anything, to do with immigrants themselves (Buck et al., 2004; Newton, 2012). Obviously, one would expect the adoption of more punitive immigration policies at a national level after the election of conservative federal politicians. Yet, our study shows that examining local administrators is equally important during these national changes in immigration opinion and salience. Local administrators, such as county sheriffs and police chiefs, enjoy a great deal of decision-making autonomy and, in many ways, are far removed from national politics. Yet, we find they are not entirely disconnected from public opinion that drives changes in federal electoral politics. If they were, we would expect to see a more random distribution of 287(g) agreements across the country. Instead, we see LEAs in conservative areas are most likely to adopt the program. Even more revealing, it is LEAs in conservative areas with large Hispanic populations or after the election of Republican presidents that are most likely to adopt the program.

Our findings further indicate these influences are interrelated. While LEAs in conservative counties are more likely to have active 287(g) MOAs, these conservative counties are even more likely to have them when they also have larger Hispanic populations and during Republican administrations. At the same time, liberal counties are far less likely to have an active MOA, regardless of the size of their Hispanic populations or who occupies the Oval Office. This suggests that future examinations of administrative discretion should not only further examine the influences of local contexts and national politics, but also the interplay between factors such as these. This would be done by focusing on the role that national-level politics plays in shaping local-level decision-making in different local-level social and political contexts. Such examinations would be especially useful considering the currently polarized national political environment and the presence of increasingly divergent policy goals of successive presidents from the two major political parties. Such examinations could provide a better understanding of how these shifting national priories shape local-level administrative behavior and how these phenomena interact at the no less politically polarized local level.

What we learn from interactions between national and local level practices in the United States can be helpful to improve our understanding elsewhere in the world, where similar political trends have had an undue influence in shaping local policies and administrative practices. Learning how nation-states exhibit similarities and differences as they perform their administrative duties at various levels of governance can be valuable to future research in comparative public administration (Oyekanmi & Agboola, 2022).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. Chy Lung v. Freeman, 92 U.S. 275 (1876).

2. 8 U.S.C. § 1325.

3. Executive Order No. 13768, 82 FR 8799 (2017–2020).

4. There does not appear to have ever been an LEA that entered into a MOA with CBP, despite the Trump Administration providing the opportunity.

5. Data for this variable were collected from archived ICE webpages. Figure 1 and the dependent variable consist only of local (i.e., non-state) MOAs.

6. We define a “southern” state as one that was of the original 11 members of the Confederacy.

7. The variable Median income is expressed in $10,000s.

8. The authors also attempted to run panel multilevel logit, Poisson, and negative binomial models with little success due to convergence and other errors in Stata.

9. We test the possibility that conservative counties also more likely to adopt 287(g) MOUs when they are confronted with another type of ethnic/racial diversity, specifically large Black populations. We do this in model B.1 in the appendix. We find the interaction term Conservatism x %Black does not reach standard levels of statistical significance.

10. The value for the Trend variable is equivalent to the number of the year for the county year, with 2006 equal to 1, 2007 equal to 2 and so on through 2020, which is equal to 15.

Appendices Appendix A

Table

Variable		Mean	Std. Dev.	Min	Max	Observations
v287g	overall	0.014	0.12	0.00	1.00	N = 47,130
	between		0.09	0.00	1.00	n = 3,142
	within		0.08	−0.92	0.95	T = 15
Conservatism	overall	0.609	0.15	0.06	0.96	N = 46,713
	between		0.14	0.09	0.94	n = 3,115
	within		0.04	0.35	0.86	T = 14.9961
%Hispanic	overall	0.088	0.13	0.00	0.97	N = 47,096
	between		0.13	0.01	0.96	n = 3,142
	within		0.01	−0.10	0.22	T = 14.9892
%Hispanic Δ	overall	0.020	0.03	−0.19	0.33	N = 47,069
	between		0.02	−0.06	0.18	n = 3,142
	within		0.01	−0.19	0.22	T bar = 14.9806
%Spanish only	overall	0.060	0.10	0.00	0.95	N = 34,528
	between		0.10	0.00	0.88	n = 3,139
	within		0.01	−0.24	0.42	T = 10.9997
Population (ln)	overall	10.262	1.48	3.81	16.13	N = 47,096
	between		1.48	4.55	16.11	n = 3,142
	within		0.04	9.26	10.91	T = 14.9892
Growth	overall	1.046	0.13	0.22	3.29	N = 47,069
	between		0.11	0.78	1.79	n = 3,142
	within		0.06	−0.19	3.27	T bar = 14.9806
%Black	overall	0.100	0.15	0.00	0.97	N = 47,087
	between		0.14	0.00	0.87	n = 3,142
	within		0.01	−0.05	0.22	T = 14.9863
Southern	overall	0.364	0.48	0.00	1.00	N = 47,130
	between		0.48	0.00	1.00	n = 3,142
	within		0.00	0.36	0.36	T = 15
Border county	overall	0.007	0.09	0.00	1.00	N = 46,965
	between		0.09	0.00	1.00	n = 3,131
	within		0.00	0.01	0.01	T = 15
LEA budget	overall	193.202	164.66	6.11	3707.45	N = 46575
	between		164.68	6.11	3707.45	n = 3105
	within		0.00	193.20	193.20	T = 15
Median income	overall	4.737	1.32	1.70	16.03	N = 47082
	between		1.20	2.18	12.50	n = 3141
	within		0.56	1.69	9.58	T bar = 14.9895
Republican pres.	overall	4.667	0.499	0	1	N = 47,130
	Between		0	4.667	4.667	n = 3,142
	within		0.499	0	1	T = 15

Appendix B

Table B1. Logit models of 287(g) counties: conservatism –% black interaction.

	Model 1
Conservatism	27.558***
	(2.987)
%Black	8.172
	(7.280)
Conservatism x %Black	9.986
	(12.557)
%Hispanic	14.529***
	(1.962)
Population (ln)	3.071***
	(0.2704)
Growth	−0.309
	(0.929)
Southern	2.979***
	(0.723)
Border county	−1.871
	(1.834)
LEA budget	0.004**
	(0.001)
Median income	1.404***
	(0.126)
Unemployment	0.390***
	(0.040)
Republican pres.	1.203***
	(0.177)
Intercept	−82.422***
	(5.381)
No. of observations	46,257
No. of counties	3,085
No. of years	15

Standard errors appear in parentheses.

***p < .001; **p < .01; *p < .05.

Appendix C

Table C1. Logit models of 287(g) counties: %Hispanic model (Model 1) with partial time-invariant variables.

	(1)	(2)	(3)	(4)	(5)	(6)
Conservatism	28.363***	28.176***	31.205***	33.343***	32.086***	28.875***
	(2.526)	(2.579)	(2.533)	(2.369)	(2.446)	(2.559)
%Hispanic	14.129***	14.716***	16.562***	17.627***	15.994***	13.903***
	(1.773)	(1.89)	(1.834)	(1.886)	(1.759)	(1.864)
Population(ln)	3.106***	3.108***	3.056***	3.257***	3.071***	3.082***
	(.264)	(.261)	(.25)	(.251)	(.244)	(.262)
Growth	.479	.472	1	1.035	.869	.318
	(.933)	(.933)	(.893)	(.902)	(.894)	(.928)
%Black	13.343***	13.23***	20.177***	21.292***	20.409***	13.374***
	(3.109)	(3.145)	(2.604)	(2.532)	(2.547)	(3.131)
Southern	3.091***	3.081***				3.049***
	(.711)	(.716)				(.721)
Border county		−1.734	−1.493	−2.156
		(1.747)	(1.768)	(2.078)
LEA budget				.004***	.004***	.004***
				(.001)	(.001)	(.002)
Median income	1.424***	1.414***	1.411***	1.448***	1.415***	1.412***
	(.122)	(.126)	(.124)	(.124)	(.12)	(.121)
Unemployment	.388***	.389***	.382***	.395***	.384***	.389***
	(.039)	(.039)	(.039)	(.04)	(.039)	(.039)
Republican pres.	1.19***	1.199***	1.152***	1.161***	1.156***	1.2***
	(.177)	(.177)	(.176)	(.179)	(.176)	(.177)
Intercept	−82.651***	−82.566***	−83.679***	−89.692***	−85.229***	−83.319***
	(4.966)	(4.982)	(4.737)	(4.527)	(4.632)	(5.091)
No. of observations	46,677	46,647	46,647	46,257	46,257	46,257
No. of counties	3,085	3,085	3,085	3,085	3,085	3,085
No. of years	15	15	15	15	15	15

Standard errors appear in parentheses.

***p < .001; **p < .01; *p < .05.

Table C2. Logit models of 287(g) counties: %Hispanic Δ model (model 2) with partial time-invariant variables.

	(7)	(8)	(9)	(10)	(11)	(12)
Conservatism	23.884***	26.48***	29.807***	30.82***	29.925***	25.871***
	(1.7)	(2.391)	(2.164)	(2.278)	(2.068)	(2.306)
%Hispanic Δ	3.805	3.935	3.985	3.543	3.613	3.818
	(4.212)	(4.302)	(4.313)	(4.301)	(4.296)	(4.244)
Population(ln)	2.943***	3.324***	3.24***	3.254***	3.175***	3.167***
	(.129)	(.276)	(.229)	(.248)	(.197)	(.25)
Growth	−.827	−.476	.303	.123	.029	−.788
	(.958)	(1.032)	(.977)	(.975)	(.977)	(1.007)
%Black	6.547**	7.724**	17.688***	17.535***	16.679***	6.688**
	(3.03)	(3.007)	(2.298)	(2.341)	(2.216)	(2.917)
Southern	4.102***	4.586***				4.439***
	(.903)	(.715)				(.699)
Border county		5.677**	6.787***	6.514***
		(2.437)	(2.207)	(2.445)
LEA budget				.006***	.006***	.005***
				(.001)	(.001)	(.001)
Median income	1.448***	1.592***	1.576***	1.566***	1.536***	1.53***
	(.117)	(.131)	(.121)	(.122)	(.119)	(.127)
Unemployment	.384***	.399***	.384***	.389***	.388***	.396***
	(.037)	(.039)	(.038)	(.039)	(.038)	(.039)
Republican pres.	1.246***	1.203***	1.141***	1.155***	1.169***	1.228***
	(.175)	(.178)	(.175)	(.176)	(.176)	(.177)
Intercept	−74.029	−82.773***	−82.862***	−84.732***	−82.719***	−80.402***
	(0)	(5.1)	(3.879)	(4.38)	(3.403)	(4.656)
No. of observations	46,677	46,647	46,647	46,257	46,257	46,257
No. of counties	3,085	3,085	3,085	3,085	3,085	3,085
No. of years	15	15	15	15	15	15

Standard errors appear in parentheses.

***p < .001; **p < .01; *p < .05.

Table C3. Logit models of 287(g) counties: conservatism - %Hispanic model (Model 3) with partial time-invariant variables.

	(13)	(14)	(15)	(16)	(17)	(18)
Conservatism	20.042***	19.485***	23.16***	24.891***	25.103***	20.708***
	(2.905)	(2.922)	(2.807)	(2.825)	(2.835)	(3.025)
%Hispanic	−13.86**	−14.373**	−12.712**	−13.589**	−13.262**	−14.499**
	(6.049)	(6.493)	(6.462)	(6.666)	(6.69)	(6.69)
Population(ln)	3.393***	3.26***	3.346***	3.545***	3.581***	3.397***
	(.27)	(.28)	(.258)	(.259)	(.259)	(.294)
Growth	−.019	−.112	.539	.635	.653	−.063
	(.975)	(.965)	(.938)	(.946)	(.948)	(.978)
%Black	11.55***	11.271***	18.902***	19.987***	20.075***	11.74***
	(3.151)	(3.074)	(2.518)	(2.497)	(2.506)	(3.241)
Southern	3.245***	3.157***				3.304***
	(.733)	(.744)				(.767)
Border county		1.82	1.316	1.246
		(2.066)	(2.281)	(2.771)
LEA budget				.004***	.004***	.004***
				(.001)	(.001)	(.001)
Median income	1.404***	1.385***	1.41***	1.469***	1.473***	1.425***
	(.125)	(.127)	(.124)	(.123)	(.123)	(.13)
Unemployment	.385***	.379***	.38***	.393***	.395***	.389***
	(.039)	(.039)	(.04)	(.04)	(.04)	(.04)
Republican president	1.18***	1.177***	1.133***	1.128***	1.13***	1.178***
	(.179)	(.179)	(.179)	(.181)	(.181)	(.181)
Conservatism x %Hispanic	46.023***	45.388***	47.36***	50.374***	50.35***	47.079***
	(9.234)	(9.716)	(9.712)	(10.005)	(10.103)	(9.969)
Intercept	−80.306***	−77.513***	−81.728***	−87.806***	−88.632***	−82.158***
	(4.85)	(5.006)	(4.619)	(4.588)	(4.572)	(5.324)
No. of observations	46,677	46,647	46,647	46,257	46,257	46,257
No. of counties	3,085	3,085	3,085	3,085	3,085	3,085
No. of years	15	15	15	15	15	15

Standard errors appear in parentheses.

***p < .001; **p < .01; *p < .05.

Table C4. Logit models of 287(g) counties: conservatism - republican pres. (Model 4) with partial time-invariant variables.

	(19)	(20)	(21)	(22)	(23)	(24)
Conservatism	22.066***	21.884***	25.293***	25.687***	25.663***	22.737***
	(2.591)	(2.766)	(2.576)	(2.652)	(2.531)	(2.69)
%Hispanic	14.405***	14.823***	16.993***	16.581***	16.07***	14.136***
	(1.881)	(1.958)	(1.871)	(1.95)	(1.792)	(1.934)
Population(ln)	3.135***	3.139***	3.155***	3.115***	3.092***	3.133***
	(.253)	(.275)	(.252)	(.257)	(.235)	(.268)
Growth	1.022	1	1.559*	1.359	1.327	.853
	(.94)	(.942)	(.889)	(.89)	(.89)	(.938)
%Black	14.097***	13.996***	21.241***	21.061***	21.013***	14.155***
	(3.157)	(3.25)	(2.626)	(2.664)	(2.584)	(3.191)
Southern	3.172***	3.128***				3.125***
	(.736)	(.736)				(.734)
Border County		−1.477	−1.553	−1.412
		(1.77)	(1.805)	(1.851)
LEA budget				.004***	.004***	.004***
				(.001)	(.001)	(.002)
Median income	1.516***	1.503***	1.515***	1.493***	1.494***	1.506***
	(.126)	(.131)	(.126)	(.127)	(.121)	(.125)
Unemployment	.391***	.392***	.388***	.389***	.386***	.393***
	(.04)	(.04)	(.04)	(.04)	(.04)	(.04)
Republican pres.	−4.047***	−4.024***	−4.095***	−4.024***	−4.017***	−3.994***
	(.701)	(.702)	(.699)	(.697)	(.694)	(.699)
Conservatism x Republican pres.	9.025***	9.002***	9.058***	8.963***	8.94***	8.959***
	(1.19)	(1.192)	(1.191)	(1.186)	(1.181)	(1.188)
Intercept	−81.204***	−80.995***	−83.527***	−83.641***	−83.159***	−82.197***
	(4.708)	(5.293)	(4.703)	(4.971)	(4.465)	(5.169)
No. of observations	46,677	46,647	46,647	46,257	46,257	46,257
No. of counties	3,085	3,085	3,085	3,085	3,085	3,085
No. of years	15	15	15	15	15	15

Standard errors appear in parentheses.

***p < .001; **p < .01; *p < .05.

Appendix D

Table D1. Logit models of 287(g) counties: Models 1, 2, 3, and 4 with year fixed effects.

	Model 1	Model 2	Model 3	Model 4
Conservatism	37.835***	37.424***	4.121	31.07***
	(4.389)	(4.39)	(5.905)	(4.412)
%Hispanic	14.363		−60.805***	−.236
	(12.481)		(16.135)	(12.435)
%Hispanic Δ		18.2***
		(5.105)
Population(ln)	5.24***	5.856***	2.97	3.151
	(2.009)	(2.099)	(2.064)	(2.113)
Growth	3.928***	2.789**	3.391***	4.176***
	(1.28)	(1.378)	(1.299)	(1.293)
%Black	29.75***	33.14***	22.436**	40.364***
	(10.687)	(10.815)	(10.942)	(10.935)
Median income	−.064	−.145	−.231	.3
	(.296)	(.297)	(.302)	(.309)
Unemployment	.146	.145	.026	.081
	(.089)	(.09)	(.093)	(.093)
Republican pres.	−3.491***	−4.561***	−4.012***	−9.21***
	(.807)	(.782)	(.838)	(1.193)
Conservatism x %Hispanic			162.631***
			(23.08)
Conservatism x Republican pres.				9.566***
				(1.413)
No. of observations	2460	2460	2460	2460
No. of counties	164	164	164	164
No. of years	15	15	15	15

Standard errors appear in parentheses.

***p < .001; **p < .01; *p < .05.

Appendix E

Table E1. Logit models of 287(g) counties: Models 1, 2, 3, and 4 with county fixed effects.

	Model 1	Model 2	Model 3	Model 4
Conservatism	41.182***	42.664***	15.106**	34.259***
	(3.73)	(3.674)	(5.033)	(3.834)
%Hispanic	54.935***		−4.851	42.977***
	(9.759)		(13.423)	(9.782)
%Hispanic Δ		6.492
		(5.192)
Population(ln)	6.934***	9.635***	5.655**	5.632**
	(1.855)	(1.92)	(1.886)	(1.909)
Growth	2.945**	1.688	2.378*	3.326**
	(1.162)	(1.29)	(1.2)	(1.18)
%Black	43.296***	61.686***	37.588***	50.970***
	(9.797)	(9.613)	(9.792)	(10.01)
Median income	1.005***	1.502***	0.929***	1.270***
	(0.212)	(0.199)	(0.219)	(0.221)
Unemployment	0.464***	0.457***	0.430***	0.454***
	(0.046)	(0.045)	(0.046)	(0.046)
Republican pres.	1.217***	1.089***	1.077***	−3.832***
	(0.207)	(0.202)	(0.214)	(0.787)
Conservatism x %Hispanic			142.759***
			(22.639)
Conservatism x Republican pres.				8.592***
				(1.312)
No. of observations	2,460	2,460	2,460	2,460
No. of counties	164	164	164	164
No. of years	15	15	15	15

Standard errors appear in parentheses.

***p < .001; **p < .01; *p < .05.

Table E2. Models 1, 2, 3, and 4 with trend variable.

	Model 1	Model 2	Model 3	Model 4
Conservatism	21.318***	17.879***	10.269***	16.933***
	(1.635)	(1.491)	(2.896)	(2.684)
%Hispanic	7.734***		−36.586***	8.803***
	(1.597)		(6.343)	(1.911)
%Hispanic Δ		21.245***
		(4.606)
Population (ln)	2.941***	2.839***	3.537***	3.125***
	(0.139)	(0.119)	(0.259)	(0.261)
Growth	4.085***	3.337***	4.461***	4.491***
	(.947)	(.902)	(.921)	(.914)
%Black	3.492	−.254	−.203	4.946
	(2.996)	(2.686)	(3.074)	(3.091)
Southern	3.054***	3.653***	3.562***	3.193***
	(0.811)	(0.723)	(0.687)	(0.659)
Border county	−1.938	1.382	2.902	−1.610
	(1.892)	(1.76)	(2.42)	(1.939)
LEA budget	0.004***	0.005***	0.005***	0.004***
	(0.001)	(0.001)	(0.001)	(0.001)
Median income	0.210	0.150	0.012	0.352**
	(0.149)	(0.145)	(0.164)	(0.159)
Unemployment	0.409***	0.402***	0.426***	0.424***
	(0.038)	(0.038)	(0.041)	(0.041)
Republican pres.	1.321***	1.225***	1.342***	−3.388***
	(0.184)	(0.185)	(0.190)	(0.718)
Trend	0.305***	0.381***	0.370***	0.292***
	(0.031)	(0.033)	(0.034)	(0.032)
Conservatism x %Hispanic			70.371***
			(9.485)
Conservatism x Republican pres.				8.21***
				(1.227)
Intercept	−74.315	−69.863	−75.203***	−76.609***
	(0)	(0)	(4.54)	(5.02)
No. of observations	46,257	46,257	46,257	46,257
No. of counties	3,085	3,085	3,085	3,085
No. of years	15	15	15	15

Standard errors appear in parentheses.

***p < .001; **p < .01; *p < .05.