How regional attitudes towards immigration shape the chance to obtain asylum: Evidence from Germany

Abstract

Asylum recognition rates in advanced democracies differ not only across states but also vary within them, translating into fluctuating individual chances to obtain protection. Existing studies on the determinants of these regional inequities typically rely on aggregate data. Utilizing a German refugee survey and leveraging a quasi-natural experiment arising from state-based allocation rules tied to national dispersal policies, we test two explanations for the perplexing regional differences. Drawing on principal–agent models of administrative decision-making, we test whether asylum decision-makers consciously or unconsciously comply with regional political preferences between 2015 and 2017 in Germany, one of the major European destination countries for refugee migration. We furthermore explore whether such biased decision-making amplifies in times of organizational stress as suggested by the statistical discrimination theory. Using mixed-effects logistic regressions, our analyses confirm a lower approval probability in regions with more immigration-averse residents or governments. We cannot confirm, however, that this association is mediated by high workloads or large knowledge gaps. Our results thus suggest that regional political biases affect the individual chance of asylum-seekers to obtain protection irrespective of temporal administrative conditions.

1. Introduction

Persecution, armed conflict, and human rights violations force people to leave their homes every day. The past decade saw the number of forcibly displaced persons grow to over 112 million by the end of 2022, strongly driven by the conflicts in Syria, Ukraine, Afghanistan, Venezuela, South Sudan, and Myanmar (UNHCR 2023). Less than a third of the displaced population seeks refuge abroad. Yet, the global dynamics of forced migration have triggered a worldwide increase in the number of asylum claims (UNHCR 2023).¹ The resulting pressure on national migration regimes (e.g. Zunes 2017) has intensified calls across receiving countries to restrict the granting of asylum. Still, almost 150 states remain parties to the two key multilateral treaties that create a binding framework for national asylum legislation: the Geneva Convention and the 1967 Protocol Relating to the Status of Refugees.

Public officials have interpreted the obligation to protect asylum-seekers differently across time and space. The people who decide on asylum requests are, to use Lipsky’s (1980) catchphrase, ‘street-level bureaucrats’ who operate in a context of constrained resources, high workloads, and ambiguous rules. The empowerment of these agents has arguably contributed to the vast divergences in recognition rates that characterize the asylum regime and translate into highly fluctuating individual chances to obtain protection.

Past research indicates that the resulting inequities are at least partly a consequence of two interrelated pressures that the street-level bureaucrats face—different regional attitudes towards migration and varying capacities to evaluate the merit of asylum applications carefully. Cross-country comparisons have identified the political preferences of the governing parties (Winn 2021) and public attitudes against immigrants (Hatton 2023) as key drivers of asylum recognition rates. The attitudes of the receiving society towards foreigners explain to some extent why the chance for asylum-seekers differs across Swiss cantons (Holzer et al. 2000a) and German states (Riedel and Schneider 2017). At the same time, the chance to obtain protection crucially depends on the volume of asylum applications currently running or filed in the past (Holzer et al. 2000b, Toshkov 2014).

The aggregate protection numbers on which these examinations typically rely, however, do not allow the researchers to assess whether attitudes towards migration and administrative capacities affect the decisions of bureaucratic agents on individual applications.² In this article, we address this problem in a detailed analysis of a representative survey of asylum-seekers in Germany. We draw on principal–agent models of administrative decision-making to identify the conditions under which street-level bureaucrats give in to the anti- or pro-migration attitudes of the population and the regional government (e.g. Holzer et al. 2000a). In light of the uncertain facts and imprecise legal standards characterizing the asylum decision (Baum 2010), we expect that the bureaucrats possess considerable discretion in the asylum domain. This facilitates asylum decisions that mirror the political preferences of the regional public and governing bodies. Our principal–agent framework thus suggests that the influence of the informal regional peers on the caseworkers explains to some extent why the chance to obtain protection varies considerably across time and space.

The administrative leeway increases further in times of organizational stress and high costs to evaluate asylum claims properly. We thus also expect that the scope for regional political preferences to shape the asylum decision expands if the available information on a case is scarce or if the administrative workload is high. These conjectures align with Lipsky’s (1980) concept of street-level bureaucracy and the statistical model of discrimination (Phelps 1972; Arrow 1973).

Based on this theoretical framework, our article investigates whether regional political preferences filter into asylum decisions in Germany, one of the primary European destination countries for asylum-seekers (Eurostat 2020), and whether information costs mediate this relationship. To test our expectations, we exploit variation across regions and conduct a detailed analysis of a unique data source, the IAB-BAMF-SOEP Survey of Refugees. The state-based allocation of refugees to the German federal states and counties upon their arrival—predominantly based on regional economic strength and population size—allows us to mitigate the refugees’ initial regional self-sorting bias (see also Aksoy et al. 2023; Kanas and Kosyakova 2023 for similar arguments). This natural experiment design enables us to minimize the challenges in the causal identification of how the regional political context shapes the individual chance for asylum-seekers to obtain protection in Germany. The survey covers a representative sample of asylum-seekers and refugees who arrived in Germany between 2013 and 2016. This data source provides detailed information on where and when the individual asylum procedure was processed and also includes a variety of individual characteristics that influence the decision on an asylum application (Kosyakova and Brücker 2020).

Using mixed-effects logistic regression models, we find strong support for our public preferences hypothesis: asylum-seekers whose applications are decided upon in regions with more immigration-averse residents or governments face a lower chance to remain in Germany as recognized refugees or as (temporarily) protected individuals. We find, however, no consistent support for the hypothesis on administrative information costs. An immigration-averse public mood does not systematically decrease the individual asylum chance further if the workload is high or if information is scarce. In sum, our analyses suggest that regional political biases affect the individual chance of asylum-seekers distributed across Germany to obtain protection. The temporal administrative conditions under which such decisions are made do not mediate these outcomes.

2. The discriminatory potential of asylum decision-making

2.1 Asylum decision-making as a principal–agent problem

The concept of asylum has existed for millennia, but it only became a global norm in the aftermath of World War I and II, when millions of people fled persecution and violence. The Geneva Refugee Convention and the Protocol Relating to the Status of Refugees have since then been translated into national legal frameworks that grant asylum to those applicants who have a well-founded fear of persecution (Joppke 1997). If this relatively strict requirement is not fulfilled, the principle of non-refoulement enshrined in the multilateral treaties forbids states to deport applicants who are likely to face harm in their country of origin.

Evaluating the claims through which asylum applicants seek to obtain protection is challenging for at least three reasons. First, any assessment of asylum claims ultimately boils down to a life-changing ‘binary yes-or-no decision’ (Taylor 2007: 176), while other administrative and judicial decisions frequently entail a range of options (e.g. the length of prison sentences).

A second challenge for decision-makers arises from the political prerogatives that they receive. The Geneva Refugee Convention and national asylum laws leave the definition of ‘political persecution’ or ‘well-founded fear’ to the decision-makers (Law 2005), granting them considerable discretion to interpret asylum legislation. This scope may, consciously or unconsciously, be used to take political preferences into account: evaluators of asylum claims may live in a region where the attitudes towards foreigners and the granting of asylum might markedly differ from the spirit of the national asylum regime and global norms. The local implementation of nationally or globally designed standards might exacerbate the principal–agent problem that characterizes policy implementation in states with a highly decentralized decision-making apparatus (Pressman and Wildavsky 1984). Specifically, the empowerment of regional agents to implement asylum law becomes problematic when street-level bureaucrats include their prejudices and stereotypes or the anti- or pro-migration attitudes of their regional peers in their asylum decisions. Such extra-legal considerations distort the asylum procedure, favouring some groups at the expense of others.

Thirdly, even the most sophisticated proceedings do not allow responsible administrators to completely overcome the informational asymmetry that they face in comparison to the applicant. Such informational barriers manifest in a lack of language skills, limited political and cultural knowledge of the asylum-seeker’s country or region of origin, and a rudimentary understanding of the applicant’s life circumstances. Hence, ‘a case may be difficult because legal standards are imprecise, the facts are uncertain, or both’ (Baum 2010: 1509). An increased workload and time pressure can exacerbate these problems.

Asylum outcomes consequently differ not only across states (Vink and Meijerink 2003, Toshkov 2014) but also within them. This could be confirmed, for instance, for Switzerland (Holzer et al. 2000a, b, Spirig 2021), the USA (Ramji-Nogales et al. 2007; Keith et al. 2013), and the UK (Jennings 2009) and is attributed to residents’ attitudes towards foreigners, the current media salience of immigration topics, the political ideology of judges, and changing workload levels, among others.

Similar variations in asylum outcomes can also be observed in Germany—the country examined in this study. Figure 1 displays the varying protection rates across federal states and countries of origin between 2015 and 2017 (also see Supplementary Table S1). For Afghans, the average protection rates range between 37 and 69 per cent, for Eritreans between 78 and 95 per cent, and for Iraqis between 40 and 93 per cent. For Syrians and applicants from the West Balkans, the divergences are more moderate, with an average protection rate of 96 per cent and 1 per cent, respectively.

Figure 1

Protection rates by German federal states and selected countries of origin, 2015–7.

Notes: Protection rates display the share of recognitions relative to all decisions. Recognitions include refugee protection, subsidiary protection, and humanitarian protection. Decisions are first-instance decisions including decisions based on the Dublin convention. West Balkan includes Albania, Bosnia-Hercegovina, Kosovo, Macedonia, Montenegro, and Serbia. For further information, refer to Table S1 in Supplementary Material. Data source: BAMF (2017), own calculations.

2.2 Regional political preferences on asylum decision-making

The federal states only possess limited discretion in asylum policy-making. Specifically, the handling of asylum claims is in the hands of the German Office for Migration and Refugees (BAMF), an agency under the jurisdiction of the Federal Ministry of the Interior. Regional BAMF offices, spread throughout the country, render verdicts on individual asylum applications.³

In the period under study, asylum-seekers were registered and assigned to federal states based on an allocation key considering the states’ tax revenues and population size (Königssteiner Schlüssel). Within the federal states, asylum-seekers were subsequently distributed by local authorities based on a similar quota logic (Geis and Orth 2016). After the initial allocation, refugees can formally apply for asylum. This first residential allocation is binding. Although housing capacities were restricted at the beginning of 2015, calculations by Gehrsitz and Ungerer (2022: 599) could not establish an ‘obvious discernible pattern in the allocation of asylum-seekers within states’. Similarly, Kanas and Kosyakova (2023) ruled out concerns that regions were able to choose refugees they wanted to host. Applications of asylum-seekers from the major countries of origin (e.g. Syria, Afghanistan, Iraq, or Eritrea) are processed in all regional BAMF offices, while some less frequent nationalities, such as Mongolia, are processed only in designated BAMF offices.⁴ No other individual features of adult asylum-seekers enter the allocation mechanism.

In their decision on an application, BAMF officials rely on the reasons given by the applicant for their application as well as on internal and external reports on the situation in the applicant’s home country and region. The interview and the asylum decision are assigned to one BAMF official only. Yet, in 2016, the process was temporarily split between one BAMF official conducting the interview and another administrator taking the asylum decision (Deutscher Bundestag 2019).⁵ Asylum-seekers can appeal a decision by the BAMF at an administrative court. Furthermore, the principle of non-refoulement implies that negative judicial decisions do not necessarily force failed applicants to leave Germany.

The delegation of decision-making power to regional BAMF branches grants its agents considerable discretion. This represents an administrative leeway through which preferences of street-level bureaucrats or their regional peers may shape asylum decisions (Jennings 2009). As we do not possess data on the decision-makers, we focus on the preferences of the state government and the public attitudes in the region in which the street-level bureaucrats operate. The salience of migration and asylum issues grew in the mid-2010s when increasing numbers of asylum requests challenged the administrative capacities of the German government. Hartland (2023) shows that higher importance attributed to migration and asylum amplified the role of immigration preferences in asylum decisions in Germany and the UK. We thus expect that the attitudes of the population with regard to immigration weigh heavily in the period under study (2015–17).

To manage the sky-rocketed numbers of asylum applications, the BAMF quadrupled its staff in this period (Grote 2018). These new decision-makers were often recruited from other units of the regional administration (Janda 2018), fortifying the regional attachment of the decision-making bodies in charge. Additionally, the BAMF case workers typically operate in tandem with other regional administrative units—in particular, with immigration offices—to coordinate the logistics of the asylum process and to gather information on the individual asylum applicant (Wittmann 2018). The expertise of the BAMF may at the same time be consulted by immigration officers (Wittmann 2018), who are in charge of immigration enforcement and the supply of social services. In this way, regional BAMF officials are regularly in touch with the regional immigration offices who are bound by the directives of their respective regional government. As a result, the individuals who decide on asylum cases are confronted with local administrative conditions and the immigration preferences of regional governments and their voters on a regular basis. Thus—consciously or subconsciously—BAMF officials might be sensitive to regional political concerns and the political agenda of the regional government when they decide on an asylum request.

Along these lines, various studies have revealed that agents’ political predispositions (e.g. Ramji-Nogales et al. 2007; Keith et al. 2013) as well as regional political preferences translate into a varying chance of a positive asylum outcome (Holzer et al. 2000a; Salehyan and Rosenblum 2008; Riedel and Schneider 2017; Schneider et al. 2020). The study by Rottman, Fariss, and Poe (2009) on asylum recognition rates in the USA before and after the 9/11 terror attacks also suggests that a changing public mood affects asylum-seekers’ chances to obtain protection. Jennings (2009) and Spirig (2021) show similarly that judicial asylum decision-making in the UK and Switzerland crucially depends on the public mood.

In line with the classic literature on implementation (Pressman and Wildavsky 1984), these findings strongly suggest that street-level bureaucrats take decisions that can contradict the federal prerogatives at least partly. We contend that the divergences from the global norm can to a certain extent be traced back to the preferences of the administrators’ informal peers. These secondary principals can be both regional government actors and citizens living in the surrounding of an asylum centre where the BAMF caseworkers decide on applications. Our theoretical framework thus suggests that the ideology of local actors shapes the federal decisions on individual asylum requests.⁶

H1: Individuals placing an asylum claim in a region with a higher share of immigration-averse residents (H1a), with a more conservative ideology of the public administration (H1b), or in a region where the state government expresses a more restrictive attitude towards refugees (H1c) have a lower chance of a positive asylum decision.

2.3 The role of information costs

The unequal treatment of asylum-seekers might also result from insufficient knowledge about the merits of an individual application. Difficulties in assessing a particular file can prompt unbiased administrators to resort to readily available information about the group to which an asylum-seeker belongs. Such stereotyping is akin to the model of statistical discrimination in labour markets developed by Arrow (1973) and Phelps (1972). Supporting this idea, Kaas and Manger’s (2012) field experiment on hiring decisions in Germany has shown that pertinent information reduces the chasm between majority and minority job applicants. Employers’ discrimination against candidates with Turkish vis-à-vis German names contracts considerably when job applications include reference letters with favourable information from previous employers.

Considerable knowledge gaps plague asylum decision-making as well. The ambiguity of both facts and law has two important consequences for asylum decision-makers. While it generates an ‘enormous leeway for choice’ (Baum 2010: 1511), it also encourages ‘motivated reasoning’ (Baum 2010: 1512, also see Kunda, 1990). As uncertainty increases, personal or public preferences for one conclusion over another gain importance in the decision-making process (Legomsky 2007). Correspondingly, BAMF officials might be more prone to regional political preferences in the absence of readily available objective evidence to support the claim. This may result in considerable regional variation in individual asylum decisions (Baum 2010).

In the case of Germany, asylum decision-makers rely on external sources such as governmental or non-governmental reports on the situation in the applicant’s country or region of origin.⁷ The need to consult external evidence is mandatory for non-refoulment decisions when the case workers have to assess the human rights situation in the country or region of origin of an applicant. An illustrative case is the overruling of the rejected asylum claim of a homosexual Palestinian asylum-seeker by the administrative court of Chemnitz. The BAMF had argued that the applicant could just move to Tel Aviv to establish a life without fear of persecution. The judges rejected this argument based on information obtained from the Foreign Office: ‘A Palestinian from Ramallah with an identity card issued by the Palestinian Authority cannot establish legal residence in Israel’ (VG Chemnitz 4 K 2610/17.A., p. 11, own translation). This example illustrates how gathering sufficient background information on a case may be critical and a failure to do so may result in the judicial revision of a BAMF decision.⁸

While we cannot assess the merit of individual asylum applications in the setup of our study, we assume along the lines of the statistical discrimination framework that the influence of regional peers on BAMF decision-makers can rise in the presence of large knowledge gaps.

H2: The absence of external sources confirming the endangerment an applicant would face in their home country reduces the chances of receiving a positive decision for applicants whose claims are decided in a region with a higher share of immigration-averse residents (H2a), with a more conservative ideology of the public administration (H2b), or in a region where the state government expresses a more restrictive attitude towards refugees (H2c).

Similarly, given that the search for externally accessible information requires time, decision-makers facing a heavy workload are more likely to rely on preferences (Andersen and Guul 2019). Previous research suggests that a stressful context affects judicial asylum decision-making (e.g. Spirig 2021). As Baum (2010) argues, decision-makers under stress are more prone to stereotyping and at a greater risk of responding ‘in terms of their general attitudes toward asylum and asylum claimants’ (Baum 2010: 1519). Time pressure could be particularly relevant in the German context, as the country faced 422,000 first-time asylum applications in 2015 and another 722,000 in 2016; almost 50 per cent of the cases submitted in Europe in this period were received by Germany (Eurostat 2020). Compared to 2014, the number of first-time asylum applications has grown by a factor of 2.6 and 4.2 in 2015 and 2016, respectively (Kosyakova and Brücker 2020).

From an administrative perspective, a large case backlog can shift the administrative focus from coherent to swift decision-making (Lipsky 1980; Baum 2010). This is what happened in Germany in the analysed period when the so-called accelerated asylum management was introduced to decrease waiting periods and to increase efficiency in asylum decision-making.⁹ We expect that in tight situations with few resources for case-specific research, decision-makers increasingly consider, consciously or unconsciously, the political preferences of their regional peers.

H3: Growing asylum workloads reduce the chances of receiving a positive decision for applicants whose claims are decided in a region with a higher share of immigration-averse residents (H3a), with a more conservative ideology of the public administration (H3b), or in a region where the state government expresses a more restrictive attitude towards refugees (H3c).

3. Data and methodology

3.1 Data

We rely on individual-level data from the longitudinal IAB-BAMF-SOEP Survey of Refugees in Germany (2016–17) (Brücker et al. 2017). The survey covers all individuals seeking asylum or any other form of protection, irrespective of their current legal status, who arrived in Germany for humanitarian reasons between 2013 and 2016 and who were officially registered by January 2017. The survey was carried out in two waves at 169 representatively selected sampling points across Germany and included responses from 7,430 adult persons. Our study targets refugees with a first-instance decision on their asylum application in the period from January 2015 to December 2017, such that the analytical sample covers 3,972 individuals (see Supplementary Material for further details on data and sample).

We augment the survey information with data on either the municipality or the federal state of the applicant’s initial residence (depending on context and data availability) and with data on their country or region of origin. Note that all regional variables are dated to the year of the asylum decision if not stated otherwise.

3.2 Outcome variables and statistical method

The dependent variable is an indicator measuring whether the asylum application has been approved (yes = 1, 0 otherwise). We do not distinguish between different forms of protection (recognition as refugee, recognition as asylum-seeker, or subsidiary protection), as we assume all forms of protection to be subject to political preferences and the public sentiments towards allowing a foreigner to stay or not. The share of applications in our sample resulting in a protection status amounts to 90.1 and 77.9 per cent in 2016 and 2017, respectively.

The unit of analysis is an individual who is nested in both the country of origin and the initial region of residence in Germany. Testing the data structure revealed (1) a high residual intraclass correlation between countries of origin and (2) a negligible residual intraclass correlation between origins within counties, given the covariates. The latter result was expected: In Germany, asylum-seekers are subject to regional state-based allocation rules according to national dispersal policies that are exogenous to the asylum-seekers’ characteristics. This assignment is regulated by the German Asylum Act and carried out by the so-called EASY system, an IT application run by the BAMF. This rules out potential biases that might have arisen as a consequence of the applicants (self-)selecting regions with better approval chances (Kosyakova and Brücker 2020). The allocation rules also reduce the potential for area effects.

To increase computational efficiency, we rely on mixed-effect logistic regression with random intercepts for countries of origin (Snijders and Bosker 1999). We decided against adding country of origin random slopes: this approach did not improve the model fit. Furthermore, for some origin countries, the data set includes a limited number of observations. We report our results as average marginal effects (AME). Full models are reported in the Supplementary Material.

3.3 Explanatory variables and hypotheses tests

Our theoretical framework implies that BAMF administrators consciously or unconsciously take the preferences of their regional peers into account when they decide on applications. To measure the political preferences of the regional residents and governments, we consider the municipality of the respondents’ initial residence. We employ five measures of the regional public attitudes present among the population, in the government of a federal state and its public administration.

To approximate the influence of immigration-averse citizens (H1a), we consider the share of residents who indicate that they are very concerned about immigration to Germany in the county of residence based on data from the German Socio-Economic Panel (SOEP 2019).¹⁰

The conservative ideology of the public administration (H1b) is measured via two proxies for the political orientation of the regional government: (1) a center-right minister indicating whether the current prime minister of a federal state belongs to a conservative party (CDU/CSU, FDP) and (2) the number of years of centre-right dominance in the federal state since 1991 through a variable assembled by Schneider et al. (2020) and the authors (see e.g. Henkes 2008, and Bultmann 2002, for a use of this proxy in studies on regional differences in naturalization decisions).

To measure the restrictive attitudes of the current regional government towards refugees in particular (H1c), we consider the application of the restrictive residence obligation by the federal states. Introduced in August 2016 but not applied by all federal states, this policy requires refugees to reside in their federal state of residence for three years after asylum approval (Wohnsitzauflage, §12a, Residence Act). German federal states furthermore decide whether or to what extent to provide benefits to asylum-seekers in kind or in vouchers instead of monetary payments (Deutscher Bundestag 2020), with a maximum amount of 143 and 216 Euros for individuals in shared accommodations and private accommodations, respectively (BGBl 2015). As vouchers for specific shops or products as well as in-kind deliveries of food, clothes, or hygienic products restrict asylum-seekers in their free choice of goods, the provision of non-monetary benefits reflects a restrictive policy as opposed to a cash pay-out. We use the yearly share of the non-monetary benefits every federal state provides to asylum-seekers to proxy state government attitudes towards refugees, with a lower share reflecting a more restrictive attitude.

Our theory also implies that a lack of time or information fuels biases in asylum decision-making. To test the hypotheses of information costs, we interact in separate models the regional characteristics with knowledge gaps on the one hand (H2a, 2b, 2c) and with the workload of asylum decision-makers on the other (H3a, 3b, 3c). To proxy for knowledge gaps, we account for the absence or existence of external sources assessing whether applicants are endangered if returned to their respective country of origin. We rely on two measures that we expect to reflect the lower bound of information available to BAMF decision-makers: (1) the Freedom House Political Rights and Civil Liberties Index (FIW), which assesses the extent to which political and civil rights are upheld in the origin countries on a yearly basis (Freedom House 2018) and (2) the log-transformed number of monthly conflict-related deaths in the region of origin (on NUTS 1 level) in the month and year of the asylum decision (Sundberg and Melander 2013; Pettersson and Öberg 2020). FIW and causalities data are widely recognized and publicly accessible. We assume that this kind of data source also informs internal government reports on the main countries of origin of asylum-seekers or resembles information therein (also see note 7). Suppose these data sources do not confirm deadly conflicts or high levels of human rights violations in the applicant’s country or region of origin. In that case, we assume a larger knowledge gap: case workers would need to run individual research on the background and the level of potential endangerment of the individual asylum applicants.

The workload of asylum decision-makers is reflected in the monthly ratio of the number of pending asylum claims to the number of decided asylum claims in the federal state of residence. The required data is published monthly (BAMF 2018).¹¹ We expect that higher administrative information costs facilitate biased asylum decision-making if the effects of regional characteristics increase with lower levels of political and civil rights violations in the country of origin (i.e. a higher FIW score), with lower numbers of conflict-related deaths in the country of origin, or with the rising workload of asylum decision-makers.

Several individual-level and contextual variables enter our analyses as potential confounders. On the side of the individual asylum-seeker, age is controlled for since it correlates with the probability of migration. Gender measured through a male dummy, religion, and pre-migration years of education are further confounders that enter the analysis. We control for traumatic experience on route (a dummy equal to 1 if such an experience was indicated and 0 otherwise) because having experienced traumatizing events on the way to Germany may result in contradictory statements and misunderstandings in the interviews with an applicant, reducing the credibility of the asylum-seeker’s testimony (Rousseau and Foxen 2010). We furthermore consider whether the respondents report that they fled their countries of origin because of violent conflict or war, forced recruitment, persecution, or discrimination. The reasons to seek refuge were pre-coded in the questionnaire, and respondents could tick multiple answers which we transformed into separate dummy variables.

To facilitate decision-making and lessen public criticism, the BAMF maintains lists of so-called ‘safe countries of origin’ and countries with ‘good perspective to stay’ (for details, see ft. 9). As the countries listed under these categories receive priority status in the processing of asylum applications, we include two respective dummy variables. A further control is whether an asylum-seeker arrived via a safe third country. If the authorities possess this information, they are likely to deny protection, partly based on the controversial Dublin regulation. We also control for population density, the share of the foreign population, and the unemployment rate at the municipality level (BBSR 2021). To account for systematic differences in the survey design data, the estimations include the sample of the survey.¹² We further control for cases where information on the application/decision dates and type was replaced with the registration date in Germany or with information provided in later survey periods. Table 1 presents the descriptive statistics for the explanatory and control variables.

Table 1

Descriptive statistics on explanatory and control variables.

Variables	Mean/share	Standard deviation	Range	Share of missings (% of sample)
Regional characteristics
Share of residents very concerned about immigration to Germany	41.33	16.24	0–100	0
Years of centre-right dominance	12.21	7.89	0–27	0
Centre-right minister	0.39		0/1	0
Application of the restrictive residence obligation	0.21		0/1	0
Share of the non-monetary benefits to asylum-seekers	51.95	17.44	0–88	0
Workload
Workload of asylum decision-makers	65.91	45.71	−0.2 to 465.3	0
Knowledge gaps
FIW score	9.77	15.68	−1 to 84	0
Conflict-related deaths	494.29	859.83	0–3,805	5.06
Controls
Fled because of forced recruitment	0.81		0/1	0.65
Fled because of persecution	0.43		0/1	0.65
Fled because of discrimination	0.50		0/1	0.65
Fled because of violent conflict or war	0.44		0/1	0.65
Male	0.61		0/1	0.00
Muslim	0.74		0/1	1.11
Years of education	9.79	5.04	0–22	5.16
Age when filing asylum application	32.03	10.49	18–87	0.76
BAMF Cluster
Good prospects to stay	0.80		0/1	0
Safe countries of origin	0.02		0/1	0
Other	0.18		0/1	0
Arrived via safe third country	0.09		0/1	0.68
Traumatic experiences on route	0.51		0/1	37.29
Population density	183.77	219.88	1–998	0
Share of foreign population	11.36	5.38	1.9–29.0	0
Unemployment rate	6.23	2.61	1.3–15.1	0
Information on application/decision dates/type was replaced	0.05		0/1	0
Sample:	0.30			0
M3			0/1
M4	0.33		0/1	0
M5	0.37		0/1	0

Notes: In the multivariate model, we control for missing values in the variables of interest.

Data source: IAB-BAMF-SOEP Survey of Refugees in Germany, 2016–7, own calculations.

4. Results

4.1 Regional differences in asylum chances

In this article, we examine in line with earlier aggregate findings on asylum decisions whether preferences of regional peers filter into the decisions on asylum applications. Our theoretical framework distinguishes three potential sources of bias: the attitude of residents towards immigration, the general ideology of the public administration, and the administration’s refugee-specific attitudes (see Fig. 2, for the full model, refer to Supplementary Table S2).

Figure 2

AME of regional characteristics on the probability of approval of the asylum application, in percentage points (with 95% confidence intervals).

Notes: Results from a mixed-effect logistic regression with random intercept for origin countries, for full model see Table S2. Data source: IAB-BAMF-SOEP Survey of Refugees in Germany, 2016-2017, own calculations.

In line with H1a, asylum applications are less successful in regions with more immigration-averse residents. A ten percentage points higher share of residents who express concerns about immigration to Germany decreases the probability of protection by one percentage point. H1b, conversely, is not empirically supported as the presence of a centre-right regional government is not significantly related to the chance of receiving protection. Also, a long-term dominance of the Christian democrats in the political history of the state increases rather than decreases the approval probability. We further find that in regions with a more restrictive administrative approach towards refugees, the chance to obtain a protection status lessens (H1c). In regions applying a restrictive residence obligation on refugees, the probability of a positive outcome is reduced by four percentage points. Likewise, an increase in the share of non-monetary benefits to asylum-seekers by ten percentage points is associated with a decrease in approval probability by one percentage point. The disparate findings on H1 suggest the absence of long-term influences of the regional government’s political tradition, highlighting instead current regional public opinion on immigration as crucial in shaping asylum decision-making: the immigration preferences of the voters and the current government are more important than the overall political tradition of the regional government in shaping the caseworkers’ asylum decisions.

Finally, screening the results for the control variables revealed per se substantially lower approval chances of Muslims and males (Supplementary Table S2), resembling findings on asylum determination in Denmark (Montgomery and Feldspang 2005) and France (Emeriau 2023). Whether these unequal approval probabilities reflect political preferences or the lower eligibility of some groups for asylum cannot be established in our analysis.

4.2 Knowledge gaps

To assess the role of knowledge gaps in biased asylum decision-making, we interacted the different facets of regional political preferences with two proxies for the absence or presence of external sources with which the possible endangerment of applicants in their home countries could be assessed. Specifically, we tested the association of political preferences with individual asylum chances at different levels of the FIW score and the number of conflict-related deaths, respectively. These two indicators proxy the lower bound of general information that decision-makers could acquire regarding the human rights situation in the asylum-seeker’s country of origin. The two operationalizations thus reflect the severity of knowledge gaps (H2a, 2b, 2c). Note that each set of hypotheses was tested in separate models (for full models, see Supplementary Tables S3 and S4).

Figures 3 and 4 depict AMEs for the characteristics of interest at different levels of the FIW score and conflict-related deaths, respectively. The graphs also report the Wald Test results for the interaction effects. We expected an additional penalty for applicants in regions with more conservative local or governmental attitudes when knowledge gaps were large. However, this cannot be confirmed. To be precise, in regions with a more immigration-averse population individual approval probabilities do not decrease further when knowledge gaps are large. Hence, H2a is not supported.

Figure 3

AME of the FIW score on the probability of approval of the asylum application, in percentage points, at different levels of FIW score (with 95% confidence intervals).

Notes: Results from mixed-effect logistic regression with random intercept for origin countries, for full models see Table S3. For each interaction effect, we performed a Wald test. To illustrate the interaction effects, we calculated the average marginal effects of the regional characteristics at different levels of the FIW score. Data source: IAB-BAMF-SOEP Survey of Refugees in Germany, 2016-2017, own calculations.

Figure 4

AME of conflict-related deaths on the probability of approval of the asylum application, in percentage points, at different levels of conflict-related deaths (with 95% confidence intervals).

Notes: Results from mixed-effect logistic regression with random intercept for origin countries, for full models see Table S4. For each interaction effect, we performed a Wald test. To illustrate the interaction effects, we calculated the average marginal effects of the regional characteristics at different levels of conflict-related deaths. Data source: IAB-BAMF-SOEP Survey of Refugees in Germany, 2016-2017, own calculations.

H2b expected growing information gaps to amplify the negative effect of a more conservative regional public administration on approval probability. We can confirm this for the political tradition of the regional administration, but not for the current government’s attitudes. Firstly, using the FIW proxy for knowledge gaps, the higher chance of obtaining asylum in regions with a longer conservative tradition decreases as information gaps grow (Fig. 3). Secondly, the penalty for applicants in regions with a current conservative government decreases and disappears with increasing knowledge gaps (using the FIW proxy). At the same time, neither of the indicators for the government’s general ideology significantly vary by knowledge gaps as measured by the number of conflict-related deaths (see the Wald test results in Fig. 4). Hence, H2b finds only some tentative support, and results are remarkably sensitive to the proxies we use.

The results also contradict H2c. The penalty for applicants in regions with a more restrictive governmental attitude towards refugees does not amplify if knowledge gaps are larger. This holds for both proxies of knowledge gaps and both proxies of the government’s attitude towards refugees (see Figs 3 and 4). Altogether, these findings suggest that contrary to the expected, regional political preferences are not more important when decision-makers have less information on an asylum-seeker’s home country. We furthermore observe a high sensitivity on the proxies applied. In other words, we cannot confirm that knowledge gaps would amplify the role of regional preferences in asylum decision-making.

4.3 Workload

We posited that a high workload and the resulting time pressure on decision-makers is a second source of biased decision-making as suggested by the theory of statistical discrimination (H3). Figure 5 illustrates whether the chance of approval declines under decision-making stress, conditional on the regional political preferences (for full models, see Supplementary Table S5). The evidence strongly suggests that time pressure does not systematically intensify the role of regional political preferences in shaping asylum decisions.

Figure 5

AME of workload on the probability of approval of the asylum application, in percentage points, at different levels of workload (with 95% confidence intervals).

Notes: Results from mixed-effect logistic regression with random intercept for origin countries, for full models see Table S5. For each interaction effect, we performed a Wald test. To illustrate the interaction effects, we calculated the average marginal effects of the regional characteristics at different levels of workload. Data source: IAB-BAMF-SOEP Survey of Refugees in Germany, 2016-2017, own calculations.

The results instead suggest that the probability of approval in regions with a more immigration-averse public does not decrease further if workloads are high. H3a thus remains without empirical support. H3b, which expected a lower approval probability in regions with a more conservative government tradition when workloads were high, is not supported either. First, the interaction effect between the workload variable and the ideological tradition of the regional government (proxied by the years of centre-right dominance) is negative, but not statistically significant. Secondly, the negative association of a current conservative government with the chance to obtain protection weakens if the workload of the caseworkers is higher. Likewise, the prediction of increased negative effects of more restrictive policies towards refugees on the chance of approval when the workload is high is not empirically supported (H3c). On the contrary, the negative association of more restrictive policies towards refugees and approval probabilities is weaker when decision-makers experience high workloads (significant interaction effect of the residence obligation indicator and no significant interaction of the non-monetary benefits indicator as suggested by the Wald test).

In sum, we cannot establish that knowledge gaps or time pressure systematically mediate the regional variation we found in the individual chance to obtain protection. Irrespective of information costs, regional attitudes towards immigration remain a constant factor in shaping the chance to obtain asylum in Germany. This highlights the considerable discretion that the street-level bureaucrats had in the implementation of the federal asylum policy in the period under study. The empirical evidence strongly suggests that the actual decision-making context does not reduce the extent to which the caseworkers are guided by the regional public mood towards immigration and asylum and that situational factors are of little importance in this context.

5. Discussion

Studies examining the varying chance of asylum-seekers to obtain protection have often faced limitations due to scarce information on the applicants’ backgrounds. Applying principal–agent models, this article breaks new ground by investigating whether the BAMF caseworkers who decide on asylum applications on behalf of the German federal state cave into the preferences of the regional population or the regional government. We furthermore examined whether this influence could be attributed to high information costs, as suggested by the theory of statistical discrimination. Specifically, we expected regional variation in the chance to obtain protection along the lines of regional political preferences to be particularly pronounced when decision-makers lacked sufficient information or time to judge the credibility of an asylum claim.

By leveraging a rich refugee survey and the state-based allocation rules linked to national dispersal policies of both federal and regional governments in Germany, our analysis provides robust insights into the complex interplay of regional political preferences and asylum decision-making outcomes. The quasi-experimental design allows us to address potential endogeneity concerns and enhances the validity of our findings. The resulting empirical evidence sheds light on the influence of regional political contexts on asylum processes, contributing to both the policy discourse and the scholarly field of migration studies.

The mixed-effect logistic regressions unambiguously demonstrated that asylum decision-making in Germany is indeed associated with the prevailing regional public attitudes towards immigration and asylum. Asylum applications were inherently less successful in regions with a more immigration-averse population or a government with a more restrictive stance towards refugees. These findings strongly suggest that the responsible caseworkers of the BAMF—an agency working on behalf of the federal government—consciously or unconsciously aligned with the current immigration preferences of their regional peers.

Following the theory of statistical discrimination, such variation in decision-making may be mediated by significant information costs. Caseworkers who face large knowledge gaps or high workloads may thus—knowingly or unknowingly—be even more prone to filter regional preferences into their decisions. Yet, our tests of the two information cost hypotheses did not find confirmation. In other words, regardless of workload or knowledge gaps, regional political preferences towards immigration and asylum consistently shaped the individual chance of asylum-seekers to obtain protection in Germany in the period under study.

In conclusion, our study confirmed that political biases enter the asylum decision-making process. However, the tests did not systematically confirm that such biases were amplified through high information costs. This indicates that the regional political sentiments that permeate the asylum decisions by the federal administrators represent a significant source of biased asylum decision-making in Germany. Global norms and national laws mandate that an asylum decision should solely be based on the merits of an individual asylum claim. Yet, our analysis provides compelling evidence that regional political preferences distort the German asylum procedure.

Our study comes with some limitations that mainly result from scarce data. We have no information on the distribution of tasks between BAMF officials or varying responsibilities of BAMF branches for different nationalities, on the number of caseworkers employed by the federal agency, and on the actual information used to assess the eligibility of an asylum claim. We thus have to rely on measures that approximate workload and information gaps. More detailed asylum statistics would not only improve research on the German asylum procedure but potentially also help the BAMF in increasing the transparency and fairness of its decision-making. As we were unable to assess the eligibility of individual asylum claims, we cannot fully rule out the possibility that unobservable factors blur our analysis. Coupling statistical analyses with interviews with applicants and administrators would be a promising avenue for future studies. Also, it should be noted that our analysis focuses on a special period in Germany’s history of immigration. In 2015 and 2016, the country experienced the largest influx of asylum-seekers since World War II, challenging local and federal administrations alike. This may have exacerbated existing deficits, such that the results should be interpreted carefully. Choosing this period of observation, however, also bears the chance to unveil existing imbalances that might go unnoticed in less stressful periods. In fact, our statistical analyses have shown that irrespective of changing workloads and information gaps, caseworkers are influenced by the regional public mood. These findings support the assumption that immigration attitudes in the surroundings of street-level bureaucrats shape asylum decisions in a way that contradicts the spirit of the global asylum regime and German federal legislation. It is beyond the scope of this article to explore the extent to which these attitudes are shaped by government ideology.

Our results strongly suggest the need for reforms on the federal level that enhance the objectivity of the asylum procedure. To counteract the influence of regional biases, caseworkers need to receive regular and uniform training in how to obtain and make use of objective information when assessing the credibility of individual asylum claims. Regular supervision sessions for BAMF officials can furthermore provide the scope to reflect, discuss, and address potential (internal and external) political biases filtering into asylum decisions. Moreover, mandating regular and systematic internal or external reviews of asylum decisions may help to limit the arbitrariness in decision-making. The systematic documentation and publication of (changing) BAMF proceedings in the asylum decision-making apparatus would represent another constituent of external control. These envisioned measures and other reforms will assist the German government in insulating asylum decision-making from the political prerogatives of the regions in which the street-level bureaucrats apply the federal asylum policy.

Acknowledgements

A previous version of this article was circulated under the title ‘Global norms, regional practices: Taste-based and statistical discrimination in German asylum decision-making’. We thank Laura Goßner, Lucas Guichard, Claus Hirsch, Alexander Kubis, Adam Scharpf, Judith Spirig, Daniel Thym, Natascha Zaun, the editor as well as two reviewers for providing helpful advice and useful comments on the earlier versions of this article, and all those who commented on the earlier presentations of this work at the following events: the XX ISA World Congress of Sociology (June/July 2023), the American Sociological Association (ASA) Virtual Annual Meeting (August 2021), the IAB-ECSR Interdisciplinary Conference ‘Refugee Migration and Integration Revisited: Lessons from the Recent Past’ conference in Nuremberg (May 2021), the Virtual European Political Science Association (EPSA) 2021 conference originally scheduled for Cologne (June 2021), the DeZIM-FG Wednesday (September 2020), and the ECSR Thematic Workshop in Milan (March 2019).

Supplementary data

Supplementary data are available at Migration Studies Journal online.

Conflict of interest statement

None declared.

Funding

This work was supported by the Federal Ministry for Family Affairs, Senior Citizens, Women and Youth (BMFSFJ, FKZ: 3920405WZB to L.G.), the Deutsche Forschungsgemeinschaft (DFG—German Research Foundation) under the Excellence Strategy of the German federal and state governments (EXC-2035/1–390681379 to G.S.) and the Institute for Employment Research (IAB, general funding to L.G. and Y.K.).

Data availability

This study uses the factually anonymous data of waves 2016–19 of the IAB-BAMF-SOEP Survey of Refugees (doi: 10.5684/soep.iab-bamf-soep-mig.2019). The survey is conducted jointly by the Institute for Employment Research (IAB), the research data center of the Federal German Office for Migration and Refugees (BAMF), and the German Socio-Economic Panel (SOEP) at the German Institute for Economic Research (DIW). Data access was provided via researchers’ contacts at the IAB. External researchers may apply for access to these data by submitting a user-contract application to the SOEP Research Data Center (https://www.diw.de/en/diw_02.c.222836.en/data_access_and_order.html). The computer codes for data preparation and analyses can be found at https://osf.io/xtm2a/ (doi: 10.17605/OSF.IO/XTM2A).

Notes

1

Our data do not cover the recent refugee flow in the aftermath of the Russian invasion of Ukraine in February 2022. Due to the European Union (EU) Member States’ activation of the so-called ‘Mass Influx Directive’, which exempts Ukrainian refugees from applying for asylum, the governance of refugee migration from Ukraine differs from other refugee groups (Brücker 2022).

2

Holzer et al. (2000a) and Spirig (2021) rely on individual-level administrative data from Switzerland, but only basic sociodemographic information on applicants was available.

3

In 2016, there were 40 regional BAMF offices (Kosyakova and Brücker 2020).

4

See Supplementary Material for robustness analyses restricted to the major asylum-seekers’ countries of origin.

5

Although these administrative particularities could differ by regional units, the BAMF does not disclose any such information that we could include in our analysis.

6

While previous studies have also stressed economic factors, the results are contradictory: Following Toshkov (2014), European recognition rates correlate positively with national income and negatively with the level of unemployment, whereas Neumayer (2005) finds no such effects. In Germany, Riedel and Schneider (2017) find a negative impact of local unemployment rates and economic growth on regional recognition rates, whereas Schneider, Segadlo, and Leue (2020) find that local GDP per capita reduces rejection rates.

7

In a personal interview with BAMF officials, we learned that such country reports exist to inform asylum decision-makers, but they were not disclosed to us.

8

The judgement was obtained from the platform Juris (Juristische Informationssystem für die Bundesrepublik Deutschland [Legal Information System for the Federal Republic of Germany]).

9

A system was introduced to cluster asylum-seekers based on the approval chances of their application (cluster A: approval rates of origin countries above 50 per cent, such as Syria or Eritrea; cluster B: safe countries of origin, such as Albania or Serbia), expected complexity of cases (cluster C: complex cases, such as Afghanistan), and travel route (cluster D: cases subject to the Dublin Agreement) (see Kosyakova and Brücker 2020).

10

We follow previous literature and consider immigration concerns to measure immigration-averse citizens (e.g. Poutvaara and Steinhardt 2018). Note that immigration concerns are also referred to as salience elsewhere (e.g. Spirig et al. 2021).

11

An even more precise measure would be the number of pending cases per caseworker. Unfortunately, the BAMF does not disclose the required information to construct this measure.

12

Controlling for decision date fixed effects (aggregated into 6-month periods) did not improve the goodness-of-model fit. Thus, we opted for a model without these controls.