Introduction

The conditions under which the police can detain deportable non-citizens, as well as the length of detention and the rights of detainees, vary among EU member states, sometimes dramatically (European Migration Network 2014). For example, in the Netherlands, non-citizens can be detained for an uninterrupted term of up to six months. This term may be applied up to three times, leading to a maximum detention term of 18 months. In France, in sharp contrast, the maximum detention term stands at 45 days. Spanish law permits the detention of deportable non-citizens for a period of up to 60 days. Because these figures imply serious consequences on the liberty of non-citizens, not only deportation but also pre-removal detention in centers for deportable non-citizens can have serious consequences on the lives of thousands. In recent years, the detention and deportation of non-citizens is increasingly being studied from different disciplines. Important studies have been conducted on the conditions in detention centers (Reference BosworthBosworth 2014), the role of NGO's in the process of detention and deportation (Reference FischerFischer 2012; Reference Kalir and WissinkKalir and Wissink 2016), the lives of deportable subjects before and after their expulsion (Reference Andrijasevic, De Genova and PeutzAndrijasevic et al. 2010; Reference KalirKalir 2010), the effect of deportability on attitudes toward state authorities (Reference Cavanagh and CauffmanCavanagh and Cauffman 2015), as well as the legal infrastructure that regulates detention and permits deportation (Reference KanstroomKanstroom 2007).

The detention and deportation of non-citizens are increasingly topping political and public debates in countries worldwide, leading some scholars to talk of a “deportation turn” (Reference GibneyGibney 2008) as part of a wider process known as the “securitization of migration” (Reference BourbeauBourbeau 2011; Reference HuysmansHuysmans 2006). This process of securitization intensifies the intersection between criminal law and immigration law, or what is now widely known in legal circles as the phenomenon of crimmigration (Reference StumpfStumpf 2011). While in recent years pre-removal detention is mostly executed in specially designated facilities and not in penitentiary ones, it still crucially involves the sanctioning of ones' fundamental right for the freedom of movement. As such, the legal decision to detain a potential deportable non-citizen calls for much prudence, not least because of the very serious social, economic and medical consequences for detainees and potentially also for their families (Reference Puthoopparambil and BjerneldPuthoopparambil and Bjerneld 2016; Reference Robjant, Hassan and KatonaRobjant et al. 2009; Reference Storm and EngbergStorm and Engberg 2013).

Like many other state projects, deportations can suffer from an implementation “deficit” or “surplus” (Reference KalirKalir 2017): they are carried out by state agents who act as “petty sovereigns” (Reference ButlerButler 2004), exercising much discretion in interpreting and applying the rules and regulations on the ground, according to their values and worldviews (Reference ChanChan 1996; Reference LipskyLipsky 1980). The case of deportation is particularly interesting, because governments may decide deliberately to delegate much power and discretion to the executive branch, and to pursue levels of implementation that are impossible to draft as formal regulations, because they are either politically controversial or in violation of international conventions and human-rights standards (Reference FeketeFekete 2005). This amplifies, in turn, the need for a judicial procedure that ensures that preremoval detention, which may lead to deportation, is applied only in cases that law specifically stipulates.

A persistent dearth in the literature on deportation processes is the study of the decisions by the various actors involved, which lead to the detention and possible deportation of noncitizens. In particular, while the issue has already been studied from a legal perspective (e.g., Reference Requejo RodríguezRequejo Rodríguez (2006) for Spain), very few empirical assessments of actual detention processes have been carried out, especially due to lack of available data. As an exception, Reference González-Beilfuss, Kalir and VallbéGonzález-Beilfuss et al. (2018) have recently undertaken an analysis of police petitions for the detention of noncitizens in the province of Barcelona, Spain. Mostly descriptive, these studies have allowed an examination of the work of all actors involved: the police, the state prosecutor, the attorney, and the judge. The authors conclude, first, that in Spain the institutions of detention and deportation of noncitizens fails according to its own standards. Second, results seem to indicate that there is a systemic discrepancy in the factors according to which the police decide on the initiation of a legal procedure for the detention of noncitizens and the criteria applied by judges in ruling on the same matter.

Decisions over preremoval detention are surrounded by high levels of uncertainty: a decision on a person's liberty must be reached within a short time relying on incomplete information—typically the one provided by court sheets or police petitions. Empirical evidence shows that, especially under these circumstances, judges and other legal actors—as boundedly rational individuals—rely on “fast and frugal heuristics” (Reference Dhami and AytonDhami and Ayton 2001; Reference Gigerenzer and GoldsteinGigerenzer and Goldstein 1996; Reference RyoRyo 2016), that is, on just a few cues of information that help them make quick decisions. The extent to which legal actors use such cues (and if so, which ones) in decisions over noncitizen detention is still a largely unexplored field.

Using data from noncitizen detention court cases in Spain, this article intends to start filling this gap. In the next section, we provide context to the Spanish case and outline the theoretical framework regarding the heuristic model of judicial decision-making. Once the framework is set, we derive a number of hypotheses from the model. Section 3 provides a description of the data set used to test the hypotheses, and results are presented in Section 4.

Theory

Context: detention of deportable noncitizens in Spain

The Spanish Immigration Act provides a specific proceeding through which noncitizens who have been issued a deportation order can be subjected to detention in centers of preremoval detention for foreign population (Centro de Internamiento de Extranjeros, CIE) for processing their forced removal from the country (Reference González-Beilfuss and Fernández-AllesGonzález-Beilfuss 2016). There are currently seven operating CIE that are run and controlled by the Spanish National Police (Reference González-Beilfuss, Arango, Mahía, Moya and Sánchez-MontijanoGonzález-Beilfuss 2017).Footnote ¹

Figure 1 depicts the proceeding that starts with the arrest of a deportable noncitizen, until a judge decides over his detention in a CIE. Once a noncitizen is arrested, everything must happen within 72 hours. The procedure starts off with the Spanish National Police Force—the only police force with the power to deal with this matter—who must decide whether they issue a petition for detention against the arrested noncitizen, in order to enforce a deportation order. If the petition is issued, a State prosecutor must proceed to provide an assessment of its legal grounds. After the prosecutor's report is issued, the petition is considered by an investigating judge who will both hear the deportable noncitizen in court, and substantiate and document a decision for or against detention. Deportable noncitizens have the right to legal counsel prior and during the hearing (and afterward if they are eventually detained in a CIE), which usually is state-appointed free legal aid. Once the hearing is finished, the judge issues a decision. If the judge denies detention, the noncitizen must be released immediately and can never be detained again for the purpose of executing the same deportation order. If authorized by the judge, detention in a CIE can last up to a maximum of 60 days. In case the noncitizen has not been removed from the country after that period (or before if deportation is deemed impossible by some circumstance), he must be released.

Figure 1. Structure of the Proceedings for Noncitizen Detention before Deportation in Spain. [Color figure can be viewed at wileyonlinelibrary.com]

According to the Spanish Immigration Act, a judge may authorize preremoval detention in a CIE according to the principle of proportionality and the concrete circumstances of the case. In particular, preremoval detention may be authorized if there is a risk of absconding, if the noncitizen hampers removal or if criminal or administrative sanctions have been imposed or may be imposed as a result of pending procedures. The lack of a fix domicile or identity documentation, the existence of criminal records and administrative sanctions, as well as of a serious illness are explicitly mentioned in the law as criteria that should be taken into account for authorizing preremoval detention. In practice, however, police records play also an important role in the whole proceeding, although this very concept raises important legal problems for various reasons. First, because it is not a legally defined concept; second, because police records usually reflect police detentions regardless of their concrete consequences (e.g., if the detainee was immediately released or if the detention was confirmed by a judge); and finally, because a police file may encompass facts that may be very different both in terms of importance and in connection with preremoval detention.

Data and method

To test these hypotheses we use a unique data set containing information of the court proceedings of all police petitions to detain deportable noncitizens in the province of Barcelona (Spain) during 2015. With 5.5 million inhabitants, the province of Barcelona is the most populated area of the Spanish region of Catalonia (7.5 million), the second most populated in Spain after Madrid (6.4 million), and one of the areas with a higher percentage of noncitizens living and working. Although Catalonia is an autonomous region within the Spanish decentralized state, judicial procedures and immigration policy are central state-dependent and homogeneous across the country. The authors were granted access to court proceedings in paper form. In order to transform them into data, a database interface was set up so that three Law Master's students could carry out the codification process of all case information in the court premises.

In 2015, 587 police petitions of noncitizen preremoval detention were admitted to court in the province of Barcelona, from which we coded data regarding: the profile of the noncitizen (age, nationality), the circumstances surrounding the arrest by the police (e.g., place of arrest), the grounds on which the police petitioned a detention in a CIE, the statements and documents presented by the State prosecutor, the statements and documents presented by the noncitizens and/or their attorneys, and the final decision and argumentation of the investigating judge.

Our main dependent variables are the decision of the prosecutor and that of the judge, which take value 0 if they decide in favor of preremoval detention in a CIE, and 1 otherwise. In our data, prosecutors decide in favor of detention 75.3% of the time (which is around the average in Spain), while judges authorize detentions in only 48.9% of police petitions.Footnote ² The unit of analysis is a case file corresponding to one single individual, and data have a hierarchical structure: cases are nested in courts within judicial districts. Although in less populated areas of Spain judicial districts may have only one court, in the province of Barcelona, districts gather several courts. Each court has one single judge, although one prosecutor attends several courts. Police petitions are brought to court to be processed by the court whose judge is on call. Petitions are then randomly assigned to a court within that judicial district.

In our data, the number of cases per court and district present high variation, with an average of 4.4 per court and a standard deviation of 3.8. The maximum number of cases per court is 22, and the minimum is 1. In addition, there is significant variation of our dependent variables across courts and districts. In order to take that variation into account, we fit linear probability models with fixed effects by judicial district.

All models include controls for the noncitizens' age and nationality. Sex is not included in the models given that an overwhelming majority of arrested noncitizens are male (96%), because the detention center of the province of Barcelona has no premises for female detainees. Should a judge order the detention of a female noncitizen, she would have to be transported to the nearest CIE having a female area, which is about 300 km away from Barcelona, and incur in additional costs. As seen in Table 1, Morocco is the most represented nationality among arrested noncitizens (39%), and 8 out the 15 nationalities with at least 10 individuals correspond to Latin America and two to sub-Saharan Africa. Some nationalities are not represented according to their actual distribution within the legal immigrant population of the province of Barcelona. Precisely, Morocco is highly overrepresented among arrested noncitizens along with Albania, Gambia, Georgia, Senegal, and Algeria. On the other hand, nationalities such as China, Ecuador, Russia, Argentina, and Peru are underrepresented. Apart from the detention process on which we focus here, the chances that a detained noncitizen is finally deported to his country after being held 60 days in a CIE are dependent on the extent to which the detainee's country of origin is willing to admit deported nationals back. Although we have no information about potential bias in arrest due to the prospects of an actual deportation (or any other factor), we cannot rule out potential bias in the data. The list of underrepresented countries contains cases known to have no agreement with the Spanish government to make deportations possible—for example, Peru and China. Regarding the overrepresented nationalities, Morocco and Algeria do have deportation agreements with Spain. However, considering that around half of those who are detained in CIEs are eventually not deported (Reference González-Beilfuss, Kalir and VallbéGonzález-Beilfuss et al. 2018), selection bias does not seem to play a crucial role. Although the effect of other social background information such as working status or social ties might have been worth testing—despite the mixed results reported in the literature (Reference DhamiDhami 2003; Reference RyoRyo 2016)—such information is not contained in the judicial files examined.

Table 1. Nationalities with at Least 10 Individuals Represented in the Data Set

To test our hypotheses, models include variables regarding legal factors and behavioral elements from prosecutors and attorneys. Regarding the former, we include the time passed between the deportation order and arrest—measured in days—in order to control for changes in immigration status. Data yield great variation, with deportation orders issued just one day before arrest while we may find noncitizens that were issued deportation orders 15 years ago (see Table 2). However, 60% of petitions refer to orders that are less than one year old, and 80% are less than 2 years old. The models also includes the cause of deportation mentioned in the court proceeding, which has been classified into three main categories that sum up the two major causes established by law: illegal stay in the country (section 53.1a of Immigration Act), and being convicted with a 1+ year prison term (section 57.2 of Immigration Act). In all, 98.4% of petitions are grounded on one of these two general causes for deportation. The rest have been coded as “other causes.” Models also control for the percentage of foreign population within a district. Foreign population in the average judicial district of the province of Barcelona constitutes around 9.8% of the total population. There are districts with very low percentages (the minimum is 3.9%) while others present number well above the mean (the maximum is 19.2%).

Table 2. Descriptive Statistics of the Continuous Variables Included in the Models

Regarding the “criminal profile” of the noncitizen, petitions contain the number of police records and the types of police records and criminal priors. Table 2 shows that the average number of police records is around nine, but with high variation. There is also variation on the types of records (police petitions mentioned more than 70 different types). When noncitizens do hold police and/or criminal records, the petitions contain a list of these records, each one identified by a short description provided by the police officer—for example, “DUI-driving under the influence (2003).” There is no minimum or maximum number of records to be mentioned in a petition, and in cases with a fairly high number of records (especially police records), officers may mention the most recent ones or the ones they consider more relevant to ground their petition. For each individual, we recorded up to five police and five criminal records—the five most recent ones—and their dates. In addition, because there is not a list of keywords or standardized descriptions to define police or criminal records shared across police offices (there is no single template for petitions, either), naming variation is extreme. We classified the types of records into six different categories.

Given that the law explicitly refers to the need that the risk that noncitizens escape a planned deportation and/or disrupt the process of deportation should be taken into account by both prosecutors and judges, a first category in our classification contains records that reflect such “risk of absconding.” Examples are pending arrest warrants, forgery of ID documentation, fraudulent marriage, obstruction of justice, escape from prison, disobedience, and resisting arrest. A second type of (police) record mentioned in police petitions refers to violations of the Immigrant Act, such as illegal stay in Spanish territory or generally the “violation of the Immigrant Act,” and they have been classified as “Immigration.” It should be noted that these are just administrative violations—precisely the ones justifying the police petition itself—and that it is in no way clear that these should count as police records. Yet, they are so in many petitions. The rest of police records have been classified according to their seriousness in Spanish criminal law: felony, lesser crime, and misdemeanor.

Table 2 also presents the distribution of each type of police record and criminal prior among arrested noncitizens. Each individual may have more than one type of police record. It is worth mentioning that among those with police records, the police mention mostly felonies. Moreover, one third of petitions with police records contain the “risk of absconding” type. Among criminal priors, felony is also first in frequency. Finally, in order to test our second set of hypotheses, we code the behavior of the prosecution and of the attorney. Prosecutors should be present in detention hearings, but sometimes they cannot attend due to lack of resources: usually one single prosecutor attends more than one court. When the prosecutor fails to attend the hearing, then judges must rely on the prosecution's written report. Our data show that prosecutors actually attend hearings 59% of the time. In the models of decision by judges we include a control for whether the prosecutor is in favor or against detention, expecting that judges will tend to agree with prosecutors. On the other hand, in Spain noncitizens have the right to legal counsel before, during and after the hearing, which usually takes the form of state-appointed attorneys. Legal counsel during hearings is compulsory, and court records contain information about the arguments used by legal counsel to defend their clients' interests in court. Our data indicate that in 20% of the cases attorneys provided no arguments whatsoever during the hearing, which reflects a passive defense strategy. In order to test the effects of the defense strategy on preremoval detention, we include a dummy variable that takes value 1 if an attorney provides active defense, and 0 otherwise.

Results

All regression tabular results in this article are presented so that negative coefficients indicate the worst-case scenario for the noncitizen (i.e., detention in a CIE), while positive coefficients indicate effects toward their release. We test our hypotheses for both the State prosecutor and the judge separately. Table 3 presents the results of a baseline model that links the decision by the State prosecution and the judge to the noncitizen's personal characteristics. Although for space reasons the coefficients of each nationality are not shown, results show that neither age nor nationality have an effect on decisions regarding preremoval detention.

Table 3. Linear Probability Regression Models of Prosecutor and Judge Decisions on Personal Characteristics of Deportable Noncitizens

Fixed effects by judicial districts.

Note: ^*p < 0.1; ^**p < 0.05; ^***p < 0.01.

These results suggest that prosecutors and judges tend instead to use legally related cues to decide over detention cases. In particular, none of the countries that are known to reject deportations (such as China, Argentina, or Peru) present significant differences compared to Morocco, which is the reference nationality and a country that accepts deported Moroccans back to its territory. Even if nationalities are grouped into broader geographic areas, none of the coefficients is significant.Footnote ³ Our first set of hypotheses deal with legal factors, and relevant results are presented in Table 4. In both models we still control for the noncitizen's personal characteristics, which continue to yield nonsignificant effects. A first result that requires our attention is that the time passed since the release of the deportation order increases the chances of the prosecutor agreeing on detention (Models 1 and 2). This runs against intuition, because there are no legal grounds on prioritizing older expulsion orders. It could be argued that longer time since the release of the order increases chances for noncitizens to be eventually arrested. For instance, a longer time since their immigration status changed could mean higher chances to accumulate police or criminal records. Besides, it could be the case that those with older deportation orders are the ones who are detained at the exit of prison—their order cannot be enforced until their sentence is served. On the contrary, longer stay could mean higher odds of being rooted in the local community and of finding more stable means of living.

Table 4. Linear Probability Regression Models of Prosecutor Decisions Including Legal and Behavioral Factors

Fixed effects by judicial district.

Note: ^*p < 0.1; ^**p < 0.05; ^***p < 0.01.

Our data seem to rule out the first line of argument, because there is only a very weak relationship between older deportation orders and the number of police records held by noncitizens (r = 0.104, p < 0.05), and those who are arrested at prison exit actually tend to have more recent deportation orders than the ones arrested in the street.Footnote ⁴ Although the effect identified is not strong, results suggest that prosecutors tend to prioritize older deportation orders over more recent ones, keeping all other things constant, but judges do not (Models 3 and 4). In neither case the place of detention has a significant effect on the decision.

Continuing with legal factors, a first difference arises between prosecutors and judges regarding the purely legal grounds of deportation. As commented earlier, the Spanish Immigration Act establishes irregular stay—living in Spain without a visa or beyond the period of voluntary exit after a visa is due—as the first ground for deportation (section 53.1.a). Beyond that, foreigners holding prison sentences for serious crimes (sentence terms of more than 1 year) are also eligible for deportation (art. 57.2). Results show that in cases when the deportation order is grounded on section 57.2 (prison sentence), the odds of the prosecution agreeing on detention are higher than when the order is grounded just on irregular stay (section 53.1.a), though the coefficient is not significantly different from zero.

In order to get a more thorough understanding of the relationship between a noncitizen's record and his chances of detention, our models also include the number of police records and the types of records cited in the police petition. Focusing on the former, Table 4 shows that the amount of police records held by a noncitizen has no significant weight into either the prosecutor's or judge's decision. However, when we look at the types of police records annotated by the police in the petition, prosecutors (Model 2) are sensitive to those that indicate certain risk of absconding, while no other types of police records or criminal priors (even holding a felony record) seem to affect the direction of the prosecution's decision. In contrast, Model 4 shows that police records indicating risk of absconding are not determinants in the judicial decision, while having police records indicating serious criminality (felony) is. Holding all other factors constant, having police records indicating felony reduces the probability of being released by 0.18 points. Having criminal records indicating felony yields also a negative coefficient, although the effects are not significant, probably due to the very reduced number of observations where particular types of criminal records are mentioned at all.

In conclusion, legal factors do affect decisions on the detention of noncitizens, but judges and prosecutors do not present a homogeneous decision-making pattern in terms of cues: while prosecutors do take risk of absconding into account, judges do not. Yet, judges take noncitizen's serious crime history into consideration (and prosecutors do not). This gives, again, only partial support to our first set of hypotheses.

We turn now to the context in which the deportation process takes place and the behavior of other actors. We focus first on the role of public prosecutors. Models 3 and 4 show that in cases when the prosecutor opposes detention, judges tend to agree, as we expected. In particular, when the prosecutor recommends that the noncitizen be released, the probability that judges acquiesce increases 0.5 points. Figure 2 shows that it actually makes all the difference: when prosecutors recommend the release of the noncitizen the probability that judges decide to release is above 0.9, while this probability would drop to 0.5 should the prosecutor recommend detention. Our hypothesis H21, therefore, receives full support from our data.

Figure 2. Predicted Probability of a Judge Releasing a Noncitizen Depending on the Direction of the Report Made by the Prosecutor.

The second factor to be considered is the role of attorneys. As commented earlier, detainees must have (usually free, state-appointed) legal counsel to attend the hearing, but sometimes attorneys only provide passive defense strategies. Results (Models 3 and 4) show that inactive counsel strategies decrease the probabilities of noncitizens being released by around 0.2 points. Figure 3 shows that the defense strategy followed by attorneys makes, again, all the difference: keeping all other factors constant, noncitizens who are actively defended by their attorneys present a predicted probability of being released above 0.6 (with all values of the 95% confidence interval above 0.5), while the point estimate of the probability of being released for those who are defended by passive lawyers is around 0.4.

Figure 3. Predicted Probability of a Judge Releasing a Noncitizen Depending on the Defense Strategy Provided by Attorney.

It might be argued that the role of attorneys might be endogenous to their perceived chances of success. In other words, attorneys might tend to be active only in cases they perceive they can win and less active where release chances look very bad for the noncitizen for “objective” reasons. However, data show that the average number of police records held by noncitizens whose attorneys provide active defense are actually a bit higher than of those whose attorneys are passive, although the differences are not statistically significant.Footnote ⁵ In addition, interacting the defense strategy with the cause of deportation, the place of arrest, or having criminal priors did yield no significant effects, either, which opens a path for further research on the determinants of defense strategy for preremoval detention that is beyond the scope of the present article. The main results found in this article are robust to a multilevel specification of the models. Multilevel models are adequate to deal with unbalanced samples at the second and third level (Reference Gelman and HillGelman and Hill 2007). In these models district-level intercepts are given a probability distribution, which relates them to the average intercept estimates are less biased than the ones we would get by fitting standard fixed-effects models. As shown in Table 5 in the Appendix, results with this more demanding specification hold. The one exception concerns the coefficient of police records indicating risk of absconding deportation in the model of the prosecutor's decision, which ceases to be statistically significant although it still has the correct sign.

Conclusions

This article offers an empirical analysis of decisions regarding preremoval detention of non-citizens. In particular, it uses data from Spanish court proceedings related to noncitizen detention to assess what factors determine decisions made by prosecutors and judges in these cases. As decisions over detention are made under conditions of uncertainty (typically time pressure and incomplete information), the article shows what cues are determinant to boundedly rational decisions.

According to this theoretical framework and previous literature on comparable settings, two sets of hypotheses were tested through linear probability models taking into account variation occurred at judicial district levels. In a first baseline model, we found that the noncitizens' personal characteristics do not have an effect on their chances of detention in centers of detention for foreign immigrants, thus aligning with previous literature (Reference DhamiDhami 2003; Reference RyoRyo 2016). In effect, neither age nor nationality have a significant weight on the decisions made by prosecutors or judges.

The first set of hypotheses referred to legal factors. According to the Spanish Immigration legislation, preremoval detention is legally justified when noncitizens present certain risk of evading or escaping their upcoming deportation if not in police custody. In addition, having a serious criminal record also makes noncitizens eligible for detention and deportation. These legal conditions led to two different hypotheses. On the one hand, we expected to find that noncitizen with records indicating risk of absconding deportation should have higher chances to be detained. On the other hand, noncitizens holding records indicating serious criminality would also have higher changes to be detained.

When legal factors were considered controlling for other variables, a first noticeable result was that the time since immigration status changed (measured as the time passed since deportation order was issued) is a significant cue for prosecutors favoring detention, although no legal grounds support it. We also found that this factor does not affect judges' decisions. We checked whether older change in immigration status is associated with higher changes to incur in criminal activities, but results ruled out this possibility.

To further test our first set of hypotheses, we added the types of police record and criminal prior held by noncitizens into the model, and results indicate that apart from the purely legal grounds of the police petition, prosecutors are also sensitive to those records that suggest risk of absconding from the noncitizen, but not to other types of police records or criminal priors (even holding a felony record) affect the direction of the prosecution's decision. In contrast, risk of escaping deportation is not a significant factor in the decisions made by judges, who instead rely heavily on the noncitizens' criminal profile. In this sense, criminal priors indicating serious criminality (felony) are a strong predictor of judges authorizing detention.

A second set of hypotheses looked at decisions in the context of other actors' behavior. In particular, we expected that judges considered the position taken by the State prosecution regarding a noncitizen's detention, and that legal counsel defense had a positive effect on the noncitizen's chances to be released. Our results give ample support to these two hypotheses. On the one hand, judges are highly sensitive to prosecutorial behavior and judges tend to agree with the prosecutor. In particular, when the prosecutor recommends releasing the noncitizen, the probability of a judge taking the same decision is very close to 1, while defendants whose detention is supported by the State prosecution have on average a probability of only 0.4 to have the judge denying detention. Turning to the role of attorneys, our results align with previous literature in finding that when attorneys do not provide effective defense, the probability of a noncitizen being released decreases significantly, while active defense strategies provide good chances for a noncitizen to be set free. These findings are consistent with previous research on immigration courts (Reference RyoRyo 2016) and suggest that restrictions on the access to legal advice in processes of detention and deportation can have serious consequences on the well-being of noncitizens. Some European countries such as Hungary, Luxembourg, Germany, Estonia, and Malta do actually restrict access to legal advice for noncitizens under certain circumstances (European Migration Network 2014).

Overall our results suggest that despite a few similarities the decision processes followed by prosecutors and judges do not conform to a common pattern of decision-making, because they tend to use different cues. Despite being experienced legal experts (due to the demanding recruiting policy adopted in Spain for both judges and prosecutors), both actors rely on just a very few cues, which resembles the pattern followed by lay magistrates in other legal systems (Reference DhamiDhami 2003). Therefore, legal expertise does not seem to provide more complex decision processes.

As an overall conclusion, regarding the main differences between judges and prosecutors, on the one hand prosecutors seem to adopt a view closer to that of the police than to the judge. Police petitions demand restrictions of liberty to assure the execution of a deportation order, and results show that prosecutors precisely decide based on cues that signal higher risk of escaping eventual deportation. In this sense, prosecutors would mostly decide to minimize that risk. On the other hand, judges do not seem especially concerned about the effectiveness with which pending deportation orders are eventually enforced. Instead, their almost unique focus on the noncitizen's felony profile indicates that they tend to authorize those detentions that they perceive minimize threats to public safety.

The results of this study, however, should be taken with caution, due to a number of limitations in our data that could only be addressed through further research and data. First of all, the reported gender unbalance in our data limits the generalization power of our models and results. This, in turn, would help exploring the potential effect of gender (both of immigrants and legal actors) in judicial behavior. Second, further information about all other actors involved (judges, prosecutors, and lawyers) might be used to refine our models and connect them to other existing models of judicial decision making that take into account judges' policy orientations and strategic behavior (Reference Epstein and KnightEpstein and Knight 2000; Reference Segal and SpaethSegal and Spaeth 2002), which would help shedding more light to a complex cognitive and behavioral process. Finally, our results also point to a third dimension of the procedure that will deserve further attention—the role of attorneys. Our data do not allow for an exploration of the factors behind the observed variation in attorney defense strategy, which limits our understanding of a procedure that, eventually, each year has serious consequences on the liberty of hundreds of people.