Equality Before the Courts? Studying Citizenship Disparities in Sentencing in Catalonia

Overrepresentation of foreigners in prisons is a well-established fact in Spain. However, there is a lack of empirical studies exploring the causes of this overrepresentation. Based on focal concerns theory, this paper aims to explore one of the possible causes of this overrepresentation: sentencing disparities between Spanish and foreign citizens. For this purpose, two subsets of court records are used to study disparities in the odds of incarceration (n = 2310) and differences in the length of prison sentence (n = 1162). Due to alleged limitations of standard regression models to analyse disparities in sentencing when differences between groups are large, entropy re-weighting was used to achieve balance between group observations with regard to control variables. The analyses reveal, on the one hand that, even when controlling for relevant legal and other extra-legal factors, citizens from Africa have higher odds than Spanish citizens to be sentenced to prison. On the other hand, no differences in the length of the conviction have been found.


Introduction
In recent decades, the foreign population 1 has been steadily increasing in Spain.Currently, it represents more than 11% of the total population (Spanish National Institute of Statistics, 2021a).This demographic change has been reflected in the prison population.Currently foreigners represent 29% of the prison population (General Secretariat of Penitentiary Institutions, 2021).Despite the growth of the foreign population, the latter percentage reflects an overrepresentation of foreign inmates because it is nearly three times higher than their percentage in the general population.These differences are more pronounced in the Catalonia Autonomous Region 2 than in the rest of Spain (see Table 1).
The purpose of this paper is to address the issue of penal overrepresentation of foreigners through the analysis of prison sentences ordered by Catalan Criminal Courts. 3The value of the work lies, on the one hand, in the lack of previous empirical research about citizenship disparities in sentencing in Spain (excluding Kemp & Varona Gómez, 2022b), and on the other hand, the innovation in the type of analysis techniques used to study differences in the odds of incarceration and in the length of the prison sentence (with the exception of MacDonald &Donnelly, 2019 andKemp &Varona Gómez &Kemp, 2020).Generally, studies about disparities in sentencing rely on standard regression techniques, which is an approach that has been criticised because of its limitations to establish treatment disparities (see a summary in Baumer, 2013).In this study, we take a step forward and apply entropy balancing to reweight observations in order to make proper comparisons.
Lastly, this criminalisation process could include disparities in sentencing.This is a recurrent object of study, with wide international evidence, reflecting possible biased application of the law (Baumer, 2013;Brandon & O'Conell, 2018).Nevertheless, in Spain, there is a lack of empirical studies on this issue.

Disparities in Sentencing Related to Sociodemographic Characteristics
Since the earlier work of Sellin (1928cited in Mustard, 2001), racial and ethnic disparities in sentencing have been a main line of criminological research in the United States (hereinafter US), whose findings usually indicate a different treatment among groups (Baumer, 2013;Johnson et al., 2010;Lee et al., 2011;Light, 2014;Light et al., 2014;Mustard, 2001;Tonry, 1995Tonry, , 2012;;Wermink et al., 2015).The reports of the US Sentencing Commission (USSC) have found that some demographical characteristics affect the sentencing outcome.For example, USCC (2017) shows that Black male defendants receive longer sentences than White male defendants, controlling for relevant legal variables.
Recently, research in the US on this matter has developed new methodological approaches.Among others, Bales & Piquero (2012) find that Blacks are more likely to be sent to prison than Whites.This result is consistent through different methods, including precision matching processing.Franklin (2015) shows direct and joint effects (with age and gender) of race and minority status in prison sentences using propensity score matching and concludes that Black and Hispanic defendants were more likely to be sentenced to prison and with longer convictions.MacDonald & Donnelly (2019), through the application of entropy weighting, do not find disparities between Blacks and Whites in the probability of being incarcerated, but they found differences in the length of the incarceration sentence.
Unfortunately, as Johnson et al., (2010) and Lee et al., (2011) pose, sentencing research is almost exclusively located in the US.Few studies about sentencing disparities have been developed in other countries.
For example, in the United Kingdom (UK), there is some research on sentencing disparities.Hopkins et al. (2016) observe ethnic disparities in the odds of incarceration for different offences in the sentences of the Crown Court, although without controlling for relevant legal factors.The report known as 'Lammy Review' (Lammy, 2017 cited in Pina Sánchez et al., 2019) introduced some control variables in the comparison of sentencing decisions between different ethnic groups and finds they are detrimental to minorities.But still some relevant circumstances were not introduced in the analyses (Pina Sánchez et al., 2019).Recently Pina Sánchez et al., (2019), relying on data scraping and text mining techniques to obtain a large sample, find no evidence of discrimination in sentencing against people with 'Muslim names' when relevant controls are introduced.Also, in the context of the UK, Pina Sánchez & Harris (2020) find that males are almost twice as likely to receive prison sentences than females, although the authors offer some possible explanations to such disparities (such as more serious previous convictions or the higher harm experienced by women in custody).
More studies of sentencing are found in other countries.For instance, Johnson et al., (2010) study sentencing disparities in homicide cases in the Netherlands.They conclude that some characteristics of the defendants influence sentencing decisions controlling for legal factors: female offenders are sentenced to less time than males and the youngest and the eldest offenders receive more lenient sentences than other age groups.Wermink et al., (2015), still in the Netherlands, confirm those results with a sample of juvenile and adult offenders.Lee et al., (2011), in their study of sentence length for drug offences in South Korea, discover that males and those who are employed or older are sentenced to longer punishment than their counterparts.In Russia, Volkov (2016) observes that for different types of offences (violent, theft and drugs) males or those who are unemployed are more likely to receive a prison sentence and, at the same time, the sentence received is more likely to be longer, while those with higher education or those who are married are less likely to be sent to prison, spending less time when they enter.

Sentencing Disparities by Citizenship Status
Some authors affirm that citizenship status had been neglected in the study of sentencing disparities (Demuth, 2002;Hartley & Armendariz, 2011;Holland, 2018;Light, 2014Light, , 2017;;Light et al., 2014;Plesničar & Kukavica, 2019;Wolfe et al., 2011;Wu & Delone, 2012).Demuth (2002) defends the necessity to add citizenship status in the sentencing research in the US for two reasons: (i) non-citizenship defendants constitute a large proportion of all defendants sentenced, and (ii) the citizenship status of defendants should not be taken into account in sentencing decisions according to US Sentencing Guidelines.
Complementarily, Wolfe et al., (2011) bring the key position of illegal immigration in politics and public policies in the last decades.All of these points are shared by Holland (2018), Light (2014Light ( , 2017) ) and Light et al., (2014) adding the key fact that criminal and immigration policies have converged, resulting in the increase of noncitizens being controlled by the states.5Furthermore, Iles & Adegun (2018) suggest that citizenship influence is even greater than other factors like race/ethnicity or even legal ones (such as criminal records or the type of offence).Lee et al., (2011), Light (2017) and Volkov (2016) state that outside the US race/ethnicity may not play the central role in the constitution of social difference.For this reason, following Light As in the case of the study of sentencing disparities in general, the vast majority of research about citizenship disparities in sentencing has taken place in the US.Within this research, it is possible to differentiate between the studies that use citizenship as one of the control variables and the studies which use citizenship as their principal explanatory variable.
In reference to the first group, generally studies find that citizenship affects the sentencing outcome in terms of the odds of incarceration.Non-US citizens are more likely to receive incarceration sentence than US citizens (Albonetti, 1997;Delone & Kautt, 2006;Johnson & Betsinger, 2009;Mustard, 2001;Ulmer, 2005).However, when it comes to the effect of citizenship on the length of the sentence, results are mixed.On the one hand, some studies point that non-US citizens receive longer sentences (Albonetti, 1997;Johnson & Betsinger, 2009;Steffensmeier & Demuth, 2000;Ulmer, 2005;USCC, 2017).On the other hand, some authors conclude that the effects of citizenship in the length of the sentence are not statistically significant (Everett & Wojtkiewicz, 2002;Kautt & Spohn, 2002;Mustard, 2001).Delving into these findings, Pasko (2002) observes that the effect of citizenship depends on the jurisdiction: non-US citizens received longer prison sentences in some regions whereas in other regions, the effect of citizenship is not relevant.On the contrary, Wu & Spohn (2010) find no effect of citizenship in the length of sentences in any jurisdiction studied.The picture in the second group of studies, which use citizenship as their principal explanatory variable, is nearly coincident.Damuth (2002) finds that non-US citizens, whether they are legal residents or not, are more likely to be incarcerated than US citizens in drug cases controlling for other relevant factors such as the criminal history or the ethnicity.Logue (2009) spots national origin within the Latino immigrant group has effects on the length of sentence.Concretely, Mexicans are treated more harshly than the rest.Wolfe et al., (2011) find direct effects of citizenship status in incarceration decisions and interactive effects of citizenship with race/ethnicity.For example, they show how Latino-US citizens receive shorter prison sentences, but Latino undocumented foreigners receive longer prison sentences.In the same direction, Wu & Delone (2012) identify shorter sentences for non-US citizens and Wu & D'Angelo (2014) reveal a negative effect of being non-US citizen in the length of sentence at an individual level (although when increases in non-citizen populations at a district level are accounted for non-US citizens received longer sentences).Light (2014), using federal sentencing data of nearly 20 years and with contextual controls, finds that non-US citizens are more likely to receive a prison sentence and longer than US citizens.Light et al., (2014: 841) conclude that citizenship is a 'powerful determinant of punishment outcomes' after finding that non-US citizens receive more sentences to prison and for more time through a series of analyses including several individual, contextual and time controls.More recently, Light (2017) detects non-US citizens, independent of their administrative status, are more likely to receive an incarceration sentence and longer than US citizens, being the Mexicans the most negatively affected.Valadez & Wang (2017) demonstrate that effects of citizenship on sentencing outcomes are moderated by other individual characteristics such as age, gender and race/ethnicity.Notably, undocumented foreigners and Hispanic male non-US citizens are the groups with higher probabilities of receiving an incarceration sentence.Within this second group, there are also specific studies by the kind of offence committed.For example, Hartley & Armendariz (2011) state that the influence of citizenship on sentence length in drug offences depends on the district: in some districts, being a non-US citizen increases the length; in others, a decrease is observed; and in others, there is no relation.In addition, Holland (2018) shows the importance of including in the study not only the citizenship status but also the nationality.Her results support the fact that non-US citizens receive worse sentencing outcomes than US citizens.In addition, among the group of non-US citizens, some receive harsher punishments, especially Mexicans.The same results are found by Iles & Adegun (2018) regarding sentence length.Complementarily, Ulmer & Parker (2019) examine different outcomes between Hispanic-US citizens and Hispanic-non-US citizens in sentence length in different locations of the US: those that are traditional Hispanic immigrant destinations and the ones that are new destinations.As the authors expected, they found there are few disparities in traditional destinations but greater disparities in the new ones.
Outside of the US few studies of the effect of citizenship on sentencing have been carried out.For example, in the Netherlands, Johnson et al., (2010) find that non-European foreigners receive longer sentences in homicide cases.In the same country, Wermink et al., (2015) detect that in general terms, non-Dutch citizens have higher odds of incarceration and receive longer sentences than Dutch nationals.In Russia, Volkov (2016) concludes Russian citizens have lower odds of receiving an incarceration sentence and when they do, they are sentenced to shorter periods.Within the German context, Light (2017) shows that non-German citizens are more likely to receive an incarceration sentence and longer ones than German citizens, with little differences between countries or regions of origin.Brandon & O'Conell (2018) in their study in Ireland detect that non-Irish nationals receive longer sentences than Irish nationals controlling for gender and previous custodial sentence.On the contrary, Plesničar & Kukavica (2019), with descriptive statistics about sentencing homicides in Slovenia, show that it seems foreigners are treated more lenient than nationals.
In Spanish criminology, there is little empirical research about sentencing.There are periodic official statistics about crime and punishments, but it is not enough to understand how courts actually work (Blay Gil & González Sánchez, 2020).Although there is no available data on people prosecuted by citizenship status, in 2015 and 2016, foreigners represented about 23% of all sentenced people (Spanish National Institute of Statistics, 2021b).With respect to regulation, to assure Courts and Tribunals follow equality in the ruling and sentencing, in the international arena, the European Convention on Human Rights (Council of Europe, 1953) assures 'Everyone is entitled to a fair and public hearing within a reasonable time by an independent and impartial tribunal established by law' (provision 6.1).However, at domestic level, Spanish legislation does not include explicitly impartiality in sentencing in any regulation, but according to the Constitutional Court, it is implicitly recognised in article 24.2 of the Spanish Constitution.
The first empirical work in Spain that addressed sentencing in Spain was Cid Moliné et al., (2002).The research issued, mainly, the legal and extralegal factors7 that lead to the final Judge's decision about suspension or substitution of a prison sentence.Although the research is not focused on citizenship, it took into account the variable nationality in the analysis.However, nationality did not present a significant effect on the final outcome.This gap in the literature calls for more research about disparities in sentencing.
More recently, other studies have taken citizenship into account to explain sentencing outcomes in specific contexts.For example, Stancu & Varona Gomez (2017) study through regression techniques the intensity of punishment imposed by Spanish Courts in sentences of murders and attempted murders.Their results show that being an immigrant is statistically significant when the model only includes socio-demographic characteristics.However, the explaining capacity of this variable disappears when adding other criminal characteristics of the case.Kemp & Varona Gómez (2022a) address the outcomes of plea bargain convictions and regarding citizenship.The authors concluded 'the results of the analyses of suspended sentences and of entering prison show that, holding other variables constant, foreign nationals are less likely to have their sentence suspended and are more likely to enter prison than Spanish defendants' (p.29).The same results are observed in Kemp & Varona Gómez (2022b).
As some authors noted (see, for example Holland, 2018;Light, 2014;Light et al., 2014;Plesničar & Kukavica, 2019), findings of citizenship's effect on sentencing are inconsistent, not so much in the case of the likelihood of being sent to prison as in the case of sentence length.Among other reasons, one explanation for inconsistency in findings could be attributed to the limited methodological approach of most of the research.Baumer (2013: 234) in his revision of the study of racial and minority disparities in sentencing identifies 'the modal approach to studying race and sentencing' (at least until the early 2010s) as the one that uses data of the adscription of the defendant to different racial/ minority groups, that sometimes uses further control variables, to predict the decision of prison sentences (and sometimes its length) with standard regression models.This is also the 'modal approach' in the case of studies analysing the impact of citizenship in sentencing outcomes (see almost all the above-mentioned studies8 : Albonetti, 1997;Demuth, 2002;Delone & Kautt, 2006;Everett & Wojtkiewicz, 2002;Hartley & Armendariz, 2011;Iles & Adegun, 2018;Johnson & Betsinger, 2009;Johnson et al., 2010;Kautt & Spohn, 2002;Light, 2017;Logue, 2009;Mustard, 2001;Pasko, 2002;Steffensmeier & Demuth, 2000;Ulmer, 2005;Valadez & Wang, 2017;Volkov, 2016;Wermink et al., 2015;Wolfe et al., 2011;Wu & Delone, 2012;and Wu & Spohn, 2010).Baumer (2013) poses, among others, two limitations to this methodology: omitted variable bias and sample selection bias.The first one occurs when relevant variables to explain the outcome are not included in the model.The second one refers to the differential probability of observations being in the sample for reasons beyond the control of researchers.These imply that estimates from the analysis are biased due to the comparison of groups that are in fact not easily comparable.Whereas the first limitation is difficult to bypass due to problems of data availability (Franklin, 2015;MacDonald & Donnelly, 2019), recent methodological developments offer possibilities to confront sample selection bias through precision matching (Bales & Piquero, 2012), propensity score matching (Franklin, 2015) or entropy balance (MacDonald & Donnelly, 2019).In this paper, we use the later to reweight observations before logistic regressions to achieve a good balance between covariates without losing effective sample size (MacDonald & Donnelly, 2019).

Theoretical Framework
Focal concern theory is one of the most used theoretical frameworks in the study of sentencing disparities (e.g.Franklin, 2015;Johnson et al., 2010;Lee et al., 2011;Light, 2017;Plesničar & Kukavica, 2019;Volkov, 2016;Wermink et al., 2015;Wolfe et al., 2011).This theory states that in order to overcome uncertainty, judges have three main concerns when making their sentencing decision: the blameworthiness of the offender, the protection of the community and the practical implications of their decisions (Steffensmeier & Demuth, 2000;Steffensmeier et al., 1998).
First, to assess the blameworthiness, judges draw on offender's culpability (from a retributive sense) and the degree of harm caused to the victim.In that assessment, the factors considered are the seriousness of the offence, the criminal history, the victimisation history (as a mitigate factor) and the concrete role played by the offender in the criminal act.Second, the protection of community concern is linked to the deterrent and incapacitating goal of the punishment.Therefore, the underpinning is the assessment of the dangerousness of the individual and the risk of reoffending.The variables that judges take into account in making such predictions are the nature of the committed offence (property, violent, drugs, etc.), the criminal history, some characteristics of the case (e.g. if a firearm was involved) and the characteristics of the offender (such as employment or the level of education).Finally, practical implications refer to the constraints and consequences of the judicial decision.For example, there are organisational concerns regarding the perspective of other criminal justice system actors, such as the prosecutor or the police, constrains about community expectations about the punishment and even concerns about the consequences of the penalty for the sentenced (in terms of breaking the bonds with their families, their health or other special needs) (see Albonetti, 1997;Steffensmeier & Demuth, 2000;Steffensmeier et al., 1998).
When confronting all these concerns, judges rarely have all the information.For this reason, they deal with uncertainty.In this situation, they must develop perceptual shorthands to fill the gaps, so judges often rely on stereotypes that link defendant characteristics such as gender, race or age with dangerousness and other concerns (see Albonetti, 1997;Steffensmeier & Demuth, 2000;Steffensmeier et al., 1998).
Some authors argue that offenders' citizenship may play a role in judicial decisions, acting as a characteristic of the offender linked with the negative public perceptions about foreigners (Hartley & Armendariz, 2011;Holland, 2018;Iles & Adegun, 2018;Johnson & Betsinger, 2009;Light et al., 2014;Light, 2014;2017;Logue, 2009;Plesničar & Kukavica, 2019;Ulmer & Parker, 2019;Valadez & Wang, 2017;Wermink et al., 2015;Wolfe et al., 2011).Although criminological research has proven foreigners are not more prone to crime than nationals (for example, Sampson, 2008;or García España, 2018), the stereotype of 'the criminal immigrant' still exists.If there is a negative opinion of immigrants within society, that may also work as a constraint for judges' decisions, even if they do not share those views (although some do as shown by Light, 2017).
Such explanations should be combined with the fact that since foreigners are a socially disadvantaged group, they have fewer resources to prevent or resist the imposition of negative labels.Consequently, the same process that places them in a vulnerable position generates a harsher treatment (Pasko, 2002;Steffensmeier & Demuth, 2002).
Furthermore, as Light (2017: 38) affirms 'given the cultural, legal, political, and normative boundaries that citizenship defines, it is likely that criminal justice actors use national membership not as a proxy for attributing negative qualities, but as an explicit status that is deserving of increased punishment', since Courts are responsible for protecting national interests.
Spanish criminological literature has shown how prejudices and stereotypes of different citizenship or national origin have built the imaginary of migration, dangerousness and crime.For instance, there is the belief of Latin-American citizens being members of criminal street gangs.In fact, the media attention and political discourse refer to them as 'Latin gangs' (Giliberti, 2014;Medina Ariza, 2010).This stereotyped opinion has also permeated to police officers, as Schmitt & Pernas (2008) found.Another group that has been linked to insecurity matters are the people from the North of Africa, especially due to the rise of islamophobia following the terrorist attacks of 11S (2001, New York) and 11 M (2004, Madrid) (Amazian, 2021;López Bargados, 2018).As Amazian (2021) stated, the racialisation of the threat and the fear was palpable in the 'war against terror'. 10More recently, the social alarm has moved towards the unaccompanied migrant children and adolescents, most of them coming from Marrakesh (42%), Mali (17.4%),Algeria (16.6%) and Senegal (13.3%) (Fiscsalía General del Estado, 2021), that are seen as a threat in some political and media discourses, generating social alarm, suspicion and rejection11 (García España & Carvalho da Silva, 2019;García España et al., 2021).In general, the African descendent population is linked to crime by media in Spain (Cea d'Ancona & Valles Martínez, 2021) and they are frequently stopped by the police (APDHA & IPAZ-UGR, 2016; García Añon et al., 2013;López Riba, 2021).The last group highlighted by the literature as being criminalised by the media and the population are Eastern Europeans.Stereotypes link these people to gangs responsible for violent burglaries and robberies.Even the police share this belief (Schmitt & Pernas, 2008) and subsequently they are more targeted in police activity (APDHA & IPAZ-UGR, 2016; García Añon et al., 2013;López Riba, 2021).
Based on this theoretical background, the hypotheses proposed for this study are: H 1 : 'Foreign people from Eastern Europe, Africa and Latin America are more likely to receive a prison sentence than Spanish citizens'.
H 2 : 'Foreign people from Eastern Europe, Africa and Latin America will receive longer prison sentences than Spanish citizens'.

Sampling
The samples used in the present work are subsets of an initial sample of 2959 case files randomly selected from Criminal Courts in the provinces of Barcelona and Girona (Catalonia Autonomous Region, Spain) during the first months of the years 2015 and 2016.The margin of error for the sample was ± 3.3% with a 95% confidence level and p = q = 0.5.
From the initial sample, some cases were excluded: (i) when the offence cannot be punished by prison 12 ; (ii) people from North America, Asia and Western Europe 13 ; (iii) naturalised individuals 14 ; and (iv) individuals older than 64. 15As a result, for studying differences in the odds of incarceration, the sample consists of 2310 cases (sample 1) whereas in the study of the differences in the length of prison sentence, the sample comprised of 1162 cases (sample 2).

Dependent Variables
Prison Sentence The variable refers to the principal conviction.It has two values: '1' if prison and '0' if other punishment (house arrest, fine, community service or deprivation of rights). 16In this case, the variable is the length of the prison sentence in months. 16It was transformed into its natural logarithm in order to reduce the original skewness.

Length of Prison Sentence
12 Because the Criminal Act does not provide prison sentences for these offences. 13Due to the insufficient sample size of these groups and their lesser perceived link to criminality. 14Due to the small size of the group. 15Due to the small size of the group.

Independent Variable
Region 17 The variable refers to the region of the country of citizenship.The coded regions are follows: 'Spain' (for nationals), 'Eastern Europe',18 'Africa'19 and 'Latin America '. 20  Control Variables Sex This variable was coded as 'Men' or 'Women'.

Age
The age variable outcomes are categorical with the following levels: under 25; from 25 to 44; from 45 to 65.

Plea Bargain 21
Two outcomes are possible for this variable: 'No PB' if there is no plea bargain and 'PB' if there is.
Recidivism 22 The variable was coded as 'Recidivism' when recidivism was considered and 'No recidivism' when it was not regarded.

Current Situation Outcomes are 'Prison' when the individual is in pretrial detention
or in prison for other causes, and 'Freedom' when the individual is not imprisoned.

Type of Offence
This variable refers to the offence for which the accused is being prosecuted.When the case involved more than one offence, the most serious offence was chosen (based on its potential linked punishment).In order to reduce the vast number of offences in the sample, a variable with broad categories was constructed.These categories are as follows: 'Theft', 'Burglary', 'Robbery', 'Injury', 'Road Safety' and 'Others'.
Table 2 and Table 3 show the basic composition of the two samples regarding the variables stated.

Analytic Strategy
In order to test the proposed hypothesis, three kinds of analysis were conducted.In the first place, a simple binary logistic or linear regression has been performed to assess the effect of the region of the country of citizenship on the odds of receiving a prison sentence and its length.As it is likely that differences exist between groups in relation with other aspects affecting sentencing decisions (such as sex, age, criminal records, type of offence) (Pina Sánchez et al., 2019), another two models of regression were estimated introducing controlling variables.
Recent works have raised some problems of standard regression models in estimating the effect of one variable in the decision making of judges due to the risk of sample selection bias and therefore biased estimates when there are important group differences across control variables23 (for example, Bales & Piquero, 2012;Franklin, 2015;MacDonald & Donnelly, 2019).For this reason, in the third analysis, more pre-processing of the data was performed to emphasise the effect of region on the odds of receiving a prison sentence and in its length while controlling for covariates.With this aim, entropy weighting24 was used to achieve balance between treatment and control group observations with regard to control variables.In this third phase, a simple binary logistic or linear regression predicting the effect of region on the odds of incarceration and in the length of the prison sentence (without control variables) has been carried out after the reweighting of control observations using entropy balance.Entropy balance is increasingly being used in sentencing research because of the tendency to compare treatment effects among groups in a non-experimental designs and its better balancing power of covariates than other methods like traditional regression, variable by variable matching or propensity score matching or weighting.MacDonald & Donnelly (2019) advocate for the use of entropy balance instead of propensity score matching or weighting to study disparities in sentencing because it achieves (quasi-)perfect balance of covariates without losing effective sample size. 25ll the analyses were performed in R Studio.For the pre-processing of the data using entropy balancing, the packages required were 'ebal' (v.0.1.6)and 'WeightIt' (v.0.10.2).

Simple Logistic/Linear Regression
On the one hand, Table 4 shows logistic regression estimates of different regions in prison sentences.These results indicate that, depending on the origin region, people have different odds of receiving an incarceration sentence in comparison with Spanish people.The difference in the odds ratio of Spanish citizens and people from Latin America 26 is not statistically significant.On the contrary, people from the other regions do have greater odds of receiving a custodial sentence than Spanish people.Foreigners from Africa are 4.46 times more likely to receive a prison sentence than Spanish defendants.In the case of people from Easter Europe, the odds ratio is 2.94.
On the other hand, in reference to the length of the prison sentence, differences between Spanish people and foreigners from Africa and Latin America ratios are low and not statistically significant.However, the estimated length of prison sentences for citizens from Eastern Europe is lower than for Spanish citizens (specifically, the estimated geometric mean of Easter Europeans is 34% lower than for Spanish people).
However, results from these regression models do not take into account differences between people from different regions regarding relevant variables such as the existence of a plea bargain or the type of offence committed.For this reason, a new logistic and linear regression models are displayed in Table 5, this time with control variables.

Multiple Logistic/Linear Regression with Control Variables
Table 5 presents the regression estimates of the effect of region and control variables on the odds of incarceration and in the prison length sentence decision.Regarding the region variable and comparing it with the estimates presented earlier, when control variables are considered, estimates for regions are different.Foreigners from Africa are significantly  Moreover, the results show other relevant variables to explain differences in the probability of being sent to prison and in the time to be served.Regarding the former, defendants with a negotiated plea bargain are 0.59 times less likely to receive a prison sentence than defendants without a plea bargain.In reference to the offence, it can be observed that the differences among the types committed are relevant in order to explain incarceration rate.People who commit injuries are 0.68 less likely to go to prison than people who commit 'other' offences.The estimated odds of receiving a prison sentence for people charged with road safety are 0.02 times lower than the odds for people having committed 'other' offences.On the contrary, the odds of receiving an incarceration sentence for people charged with theft are 2.53 times higher than the odds of people committing 'other' offences.In reference to the length of the sentence, when criminal records are considered as recidivism, the estimated duration is longer.Compared to being processed from freedom, defendants in prison tend to receive longer sentences.Finally, concerning the type of crime, in comparison with people who have committed 'other' offences, those who have committed road safety offences or theft tend to receive shorter prison sentences, while those who have committed robbery or burglary tend to receive longer ones.
As stated before, even when controlling for other covariates, if differences between people of different regions are pronounced, it is likely that regression estimates could be biased.In order to bypass this limitation, pre-processing of the data using entropy balance was carried out (results of this balancing or re-weighting process can be seen in the plots displayed in the Annexes Figs. 1 and 2).

Weighted Logistic/Linear Regression
Once the adjusting of observations using entropy balance process was carried out, new logistic and linear regression models were constructed only with the region variable.Table 6 shows the estimates of this regression.Estimates from the logistic regression show that people from Africa face higher odds of incarceration than Spanish nationals (1.35).No statistically significant differences are found for the rest of regions.On the other hand, when estimating the effects of region on the length of the prison sentence, no statistically significant differences are detected.

Discussion
Some scholars defend citizenship plays a role in the construction of social difference and that it is a mechanism of stratification.For this reason, some of them considered that citizenship can be involved in judicial decisions (Hartley & Armendariz, 2011;Holland, 2018;Iles & Adegun, 2018;Johnson & Betsinger, 2009;Light et al., 2014;Light, 2014;2017;Logue, 2009; are culturally closer compared to other groups of foreigners. 28The same could be said about people from Eastern Europe (even if the first model points out that they face higher odds of incarceration, the next two models show there is no statistically significant differences with respect to Spanish citizens).Conversely, there is no evidence to confirm the second hypothesis.Our analyses suggest there are no statistically significant differences in the length of the prison sentence.In fact, the first two analyses pointed to a lenient treatment for people from Eastern Europe.However, this effect disappears in the last and more robust analysis.
Examining the rest of the findings under the light of the focal concerns theory, we found evidence that both legal and extralegal factors influence prison sentence decisions.The only variable related to blameworthiness of the defendant with a significant effect on the final conviction is the type of offence.In comparison with the category of 'other' offences, having committed a theft increases the likelihood of being convicted to prison whereas having committed a road safety offence reduces the odds of being sent to prison.Concerning the length of the sentence, having committed theft or a road safety offence predicted shorter sentences as compared to other offences while having committed robbery or burglary predicted longer ones.Under the focal concern theory, this is a proxy of the blameworthiness concern, because it may be said that robbery and burglary are more serious offences than theft and road safety ones..The variables connected to the will of protecting the community are the consideration of criminal records as recidivism.This factor does not affect independently the odds of receiving a prison sentence but predicted longer ones, probably this is due to the retribution aim of the punishment.Finally, the achievement of a plea bargain could be considered as an indicator of the practical consequences that Criminal Courts take into account.When there is a plea bargain agreement, the odds of receiving a prison sentence are reduced (but the same factor does not have a significant effect in the length of the prison sentence).This could be explained because when there is a plea bargain, the odds of having the prison sentence suspended increase (see Varona Gómez & Kemp, 2020).

Limitations
Compared to previous studies of sentencing decision-making process, our research has a number of strengths.However, it has some limitations that merit discussion.In the first place, the data comes from two different jurisdictions, Barcelona which is a big province, and Girona, a smaller one.Moreover, the data was collected from Criminal Courts, regardless that other kinds of courts could impose prison for different offences.For these reasons, more research is also needed to assess if the findings in these Courts can be generalised to decisions in other locations and other kinds of Courts.Secondly, there might be omitted variable bias impossible to overcome due to (a lack of) availability of data (Franklin, 2015;MacDonald & Donnelly, 2019).For instance, variables related to defendant characteristics such as socio-economic status (as has been stated by Baumer, 2013;Franklin, 2015;or Volkov, 2016) or the educational background, the role played by the prosecution and even the characteristics of the judges themselves (in this sense see Bales & Piquero, 2012).Thirdly, the categorisation of the independent variable contains vast regions, such Africa, which includes so many different countries.Fourthly, as focal concerns theory defends, there might be interactions among variables that we did not consider properly with the techniques used (Steffensmeier et al., 1998).Finally, the sample of women is small, and this makes it difficult to find differences by sex.We recommend further studies in the future considering intersectional theories focused on gender (as stressed by Vasilescu, 2019).

Final Remarks
In conclusion, our findings show that legal and extralegal factors affect punishment outcomes of Barcelona and Girona Criminal Courts.In terms of citizenship, the general trend is that citizens from Africa have higher odds than Spanish citizens of being sentenced to prison.The results can be explained by the focal concerns perspective.We defend that the stereotypes attributed to criminalise people from Africa make them seen as a threat to social order that should be controlled using all the available means -including the Criminal Justice System.However, once they are sentenced to prison, it seems there are no differences in the length of the conviction.
As Plesničar & Kukavica (2019) show this is only one aspect of the multifaceted issue of 'crimmigration'.Nevertheless, it is still important because equality before the law is a key value in the rule of law, and disparities in sentencing threaten this value and undermine trust in the Criminal Justice System (Brandon & O'Conell, 2018;Steffensmeier & Demuth, 2000).
Further research is needed to address the issue of discrimination.We acknowledge disparities can imply discrimination, but they can also occur for different reasons such as a previous criminal selection process (Baumer, 2013).Current research cannot conclude that some defendants have been discriminated against, we can only point out that there are disparities in sentencing.The fact is that is hard to prove discrimination through statistical analyses.For this reason, we encourage qualitative research on this matter, such as the one conducted by Light (2017).

Table 5
Regression estimates of the effect of region on incarceration sentence decision and in the length of the sentence to prison with control variables *p < 0.05; **p < 0.01; ***p < 0.001 a The very high regression coefficients for burglary and robbery are due to the fact that all persons of the sample who were convicted of these offences received a prison sentence.

Table 6
Regression estimates of the effect of region on incarceration sentence decision and in the length of the sentence to prison with re-weighting of observations *p < 0.05; **p < 0.01; ***p < 0.001 AnnexesFig. 1 Absolute mean differences of covariates before and after entropy balance in sample 1 Fig. 2 Absolute mean differences of covariates before and after entropy balance in sample 2