Scared Straight?Threat and Assimilation of Refugees in Germany∗

Philipp Jaschke† Sulin Sardoschau‡ Marco Tabellini§

March 2021

Abstract

This paper examines the role of threat and locals’ hostility in determining theassimilation path of immigrants, exploiting plausibly exogenous variation in theallocation of refugees across German regions between 2013 and 2018. We assemblea novel data set on values, habits, and preferences, as well as economic outcomesfor 8,000 refugees, and combine it with information on more than 34,000 locals.We distinguish between one-sided assimilation effort (signaling cultural similarityin stated preferences) and cooperative assimilation success (labor market integra-tion). We find strong evidence that refugees who were allocated to regions thatsupport anti-immigrant parties and commit more hate-crimes against refugees, ex-ert more assimilation effort but are less successful in assimilating. We take thisas evidence for what we label the “threat hypothesis” of assimilation. We providean array of robustness and plausibility checks, ruling out ext-ante selection on theside of authorities and ex-post sorting on the side of refugees as drivers of ourresults. Lastly, we examine the role of refugee networks and the distribution oflocals’ preferences.

Keywords: Migration, refugees, integration, assimilation, identity, threatJEL classification: F22, Z10

∗We thank participants of The Economics of Migration Senior Seminar, the IAB Colloquium, as well and the 10thAnnual Conference on Immigration in OECD countries for their helpful comments and suggestions. Michelle Harnischprovided outstanding research assistance. All remaining errors are ours.

†Institute for Employment Research (IAB). Email: Philipp.Jaschke@iab.de‡Humboldt University, Department of Economics. Email: sulin.sardoschau@hu-berlin.de§Harvard Business School, CEPR, and IZA. Email: mtabellini@hbs.edu

1 Introduction

The dramatic increase in international migration flows has put the issue of immigrant

assimilation at the forefront of the political debate. Adding to the movement of hundreds

of millions of economic migrants, recent years have witnessed the unprecedented rise

in the number of refugees (United Nations, 2018). A recurring theme underlying the

public discourse around the refugee crisis – both in the global North and South – is

the concern that refugees are not able or willing to assimilate in host societies. This

debate frequently centers around refugees’ attributes (and supposed lack thereof), such

as education, language skills, religious habits, values, and norms.

The underlying logic of many integration policies is that assimilation pressure is

needed to guarantee refugees’ successful assimilation. In recent years, several European

countries have imposed restrictions on dressing habits of Muslim women – a policy that

is supported by a substantial proportion of citizens.1 In 2016, Germany passed the Inte-

gration Act, which prohibited the free movement of refugees for fear of “refugee ghettos”.

Conversely, the role of the host society, and specifically the behavior and attitudes of

locals, is rarely viewed as an important determinant of refugees’ integration.2 This in-

cludes the above mentioned policies and their unintended consequences (Abdelgadir &

Fouka, 2020; Fouka, 2020; Rumbaut, 2008), but also extends to threat and assimilation

pressure exerted, directly or indirectly, by citizens at the local level.

The goal of this paper is to examine the role of threat and locals’ attitudes in de-

termining the assimilation trajectory of refugees. We relate to the growing literature

in economics and political science on the effects of threat and pressure on immigrants’

assimilation. This literature finds ambiguous results. On the one hand, a more friendly

environment might make it easier for refugees to integrate by facilitating cross-group

interactions. Similarly, lack of openness and forced assimilation may trigger backlash

among immigrants, who try to preserve their own cultural norms (Abdelgadir & Fouka,

2020; Fouka, 2020). On the other hand, natives’ opposition to refugees may heighten

incentives to signal allegiance to the nation and its values – a process we label “threat hy-

pothesis”, and emphasized in previous work (Fouka, 2019; Saavedra, 2018; Fouka et al.,

2021; Bisin & Tura, 2019).

We begin by presenting a simple conceptual framework that links threat to assim-

1For instance, a majority of European voters are in favor of introducing bans on Islamic veils. See PEW Report andAbdelgadir & Fouka (2020).

2We use the term “refugee” for a local resident applying for or having received asylum status and “local” as short-hand for a local resident living in the same NUTS2 region, but not applying for or having received asylum status. Weacknowledge that both refugees and non-refugees are locals, and that the terms used are simplifying and imprecise.

1

ilation, and provides testable implications, which we can bring to the data. Similar

to Fouka et al. (2021), our framework distinguishes between assimilation effort and as-

similation success. We define the former as an input factor into successful assimilation

that refugees can undertake irrespective of locals’ actions. Instead, we view assimilation

success as an equilibrium outcome that depends both on locals’ willingness to interact

with outgroup members and on the effort exerted by minoritized groups. Specifically, we

build on the intuition that refugees need to exert costly effort to adopt (or to signal) local

habits, values, and norms in order to access social and economic opportunities. How-

ever, in order for refugees to successfully assimilate, locals have to be willing to interact

with them. When interacting with strangers (i.e. refugees), locals pay a psychological

cost, which is decreasing in the (actual or perceived) cultural similarity between groups.

In places with more hostile attitudes against minoritized groups – areas that we label

as “threat environments” – locals demand higher levels of cultural similarity to accept

refugees as part of the society.

This simple framework delivers three predictions. First, effort, but not successful

assimilation, should be higher in more threatening environments (as long as refugees

choose to exert any effort at all).3 Second, in more hostile environments, effort, but

not successful assimilation, may display non-linearities, as refugees decide to exert more

effort early on. Third, by lowering the pressure faced by refugees to assimilate and

possibly reducing the perception of threat, larger ethnic networks should be associated

with lower effort, while their impact on successful assimilation is ambiguous.

In contrast with most existing studies, we are able to estimate the short-run effects

of threat on both effort and successful assimilation. We consider the case of Germany,

which received more than 1.6 million refugees between 2013 and 2018. We measure

assimilation effort with preferences, values, and norms reported by refugees in a novel

survey dataset that follows the same individual over time. We instead use refugees’ labor

market outcomes, which depend on both refugees’ actions and locals’ behavior, to proxy

for successful assimilation.

Our analysis relies on two main datasets. First, we use the novel IAB-BAMF-SOEP

Survey of Refugees – a longitudinal and nationally representative survey that collects in-

formation on socio-demographic characteristics as well as values, habits, and preferences

for around 8,000 refugees. Second, we measure preferences and values for more than

30,000 locals from the German Socio-Economic Panel. We define cultural similarity in

3The ultimate effect of the “threat environment” on refugees’ successful assimilation is ambiguous. On the one hand,more hostile places are characterized by higher discrimination. On the other, in these areas refugees exert more effort andthus converge to locals’ habits faster.

2

stated preferences at the individual (refugee) level by computing the distance between

the answer provided by individual refugees on a variety of questions about social and

cultural preferences and the answer to the very same questions given by all “locals” in

the years prior to the influx of refugees.4

We focus on local, rather than on national, determinants of assimilation for two

reasons. First, existing evidence suggests that individuals are more concerned about

the effects that refugees and immigrants have on local communities and their values,

rather than on the country as a whole (Card et al., 2012; Enos, 2014; Halla et al., 2017).

Second, threat perceptions are more likely to arise in response to direct experiences

(Paterson & Neufeld, 1987). While refugees may update beliefs on hostility based on

national news coverage and integration policies, local factors, such as support for far

right parties or hate crimes, are likely to be more powerful in increasing the salience of

threat and hostility.

We exploit refugees’ quasi-random allocation across German regions between 2013

and 2016, and control for a wide range of individual characteristics (e.g. gender, age,

partnership status, education, and work experience upon arrival) as well as for NUTS-2

region and district fixed effects.5 First, we examine how refugees’ assimilation effort

(stated preferences) and success (self-reported employment status) change with each

additional month spent in the region of assignment. Specifically, we use an Intent to

Treat (ITT) approach, measuring both refugees’ outcomes and the local environment

(e.g. culture and the level of threat) in the region of assignment of the refugee, instead

of the region of actual residence. We find that, every year, the “cultural gap” between

refugees and locals living in the same region decreases by 5%, while the probability of

being employed among refugees increases by almost 50% relative to the sample mean.

Next, we estimate the impact of local-level threat on both assimilation effort and

assimilation success. We measure the prevalence of local threat with support for far

right parties and with the number of hate crimes committed against refugees. To obtain

a unique proxy for local threat, we combine these variables into a unique index using its

first principal component (PCA). We document that refugees increase their assimilation

effort more in areas where the local environment is more hostile. These patterns are

evident especially early on, as refugees become “culturally closer” to locals during the

4We consider a variety of cultural and socioeconomic preferences, such as risk preferences, importance of leisure, andreciprocity. Building on Cha (2007), we construct an index of Euclidean distance to measure refugees’ convergence to localculture over time. Intuitively, the index captures the shortest, unweighted distance between two points in the culturalspace. We describe the construction of the index in more detail below.

5Districts in Germany correspond to NUTS-3 regions, and are called Kreisfreie Stadte and Landkreise. There are morethan 400 districts in Germany with an average number of 180,000 inhabitants.

3

first 12 to 24 months since arrival. Our estimates are quantitatively large. Comparing

a refugee allocated to a region at the 75th percentile of the distribution of the “threat

index” to one allocated to a region at the 25th percentile, the former converges 50% faster

to locals’ average culture relative to the latter. However, consistent with higher threat

being associated with stronger discrimination, higher effort does not map into higher

assimilation success. In fact, even though our estimates are imprecise, they imply that,

if anything, refugees’ employment growth is slower in more hostile areas.

We provide evidence that our results are not driven either by ex-ante selection on

the side of authorities or by ex-post sorting on the side of refugees. In particular, we

show that our findings are unlikely to be influenced by: i) changes in the composition of

refugees – e.g. with individuals who are more likely to converge towards local culture or

economically integrate to move to Germany over time; ii) changes in assignment policies

over time – e.g. refugees being assigned to places with different cultural and economic

characteristics and threat levels; iii) selective internal migration – e.g. with refugees

relocating to areas that are a better cultural or economic match for them or differential

out-migration from threat regions; and, iv) underlying economic factors that correlate

with the threat environment.

In the second part of the paper, we present additional results that are consistent with

the “threat hypothesis”. First, we find that refugees converge faster in regions where

cultural assimilation pressure – as measured by the distinctiveness of locals culture and

its internal homogeneity – is higher.6 Second, we exploit the random assignment (given

pre-entry characteristics) of refugees to different types of accommodations (centralized

refugee housing versus decentralized social housing) to investigate threat perceptions at

the individual, rather than at the regional, level. We hypothesize that the individual-

level threat perception should be higher when refugees are easily identifiable (e.g., when

they live in known refugee shelters), more so in “high threat” regions. In line with this

conjecture, we show that refugees increase assimilation effort more when they live in

refugee shelters that are located in higher threat regions.

We conclude by examining the effect of ethnic enclaves on both assimilation effort

and assimilation success. We proxy for networks using either the refugee share in the

region or the average employment outcomes of immigrants (from the same origin) at

baseline. Interacting months since arrival with either proxy, we find that larger networks

slow down assimilation effort, but have no effect on either employment or wages. The

6An alternative possibility, not necessarily in contrast with our preferred one, is that it might be easier for refugees tounderstand what the “relevant” social norms are in places where local culture is more distinct.

4

negative effect on effort, however, disappears when focusing on the average labor market

outcomes of refugees in the pre-period.7 These patterns are consistent with the existing

literature. In line with previous work (Eriksson, 2020; Lazear, 1999), when refugees

live in larger networks, they have lower incentives to assimilate – possibly also because

they perceive a lower level of threat. At the same time, a larger network may make it

easier to find a job (Battisti et al., 2016; Edin et al., 2003); this positive network effect

on employment, however, may be counterbalanced by the lower assimilation effort and,

possibly, by higher labor market competition due to the presence of workers that are

closer substitutes.

Our paper is related to different strands of the literature. First, we contribute to

the broader literature on cultural transmission. In the context of migration, economists

have analyzed immigration-induced changes in preferences of natives (Alesina et al.,

2018; Dahlberg et al., 2012; Giuliano & Tabellini, 2020; Steinmayr, 2020), the influence

of emigrants on the cultural dynamics of the origin community (Barsbai et al., 2017;

Rapoport et al., 2020), and changes in or the persistence of immigrants’ preferences

(Abramitzky et al., 2020; Fernandez & Fogli, 2009). Most closely related to our paper,

Abramitzky et al. (2020) show that both today and in the past, immigrants gradually

assimilate culturally in the United States. Our estimates are quantitatively larger than

those documented by Abramitzky et al. (2020) in the US context, both for the early

twentieth century and for the more recent period. This may be for at least two reasons.

First, we consider a more direct proxy (though not necessarily superior) for culture,

constructed directly from the revealed preferences of natives and refugees. Instead,

Abramitzky et al. (2020) use names chosen by immigrants for their kids to measure

cultural convergence. Second, we focus on local, rather than on national, culture, as

done instead in Abramitzky et al. (2020).

In previous work, Abramitzky et al. (2014) have documented that European im-

migrants did not always upgrade fast economically in the first half of the twentieth

century. Findings in Fouka et al. (2021) suggest that slow economic convergence was at

least partly due to social (and economic) discrimination from natives, which fell once

African Americans arrived to cities in the US North between 1910 and 1930, as the

racial classification got re-defined around the color line. We contribute to this literature

by focusing on local, rather than national, culture and by studying the experience of

refugees rather than economic migrants.

7It is possible that, when living in areas with more successful peers (as proxied by their labor market outcomes), arefugee has higher incentives to exert effort, because she believes success is more likely. That is, more successful peersmay become “role models”, leading to higher incentives to exert effort.

5

Second, our paper is related to the vast and growing literature on the economic in-

tegration of refugees in high-income countries, which has recently been summarized in

Brell et al. (2020) and Becker & Ferrara (2019) among others. We complement these

works by focusing on cultural convergence, and by bringing in a novel dataset that allows

us to track refugees’ cultural and social preferences over time. Since most interactions

occur at the level of communities, focusing on this geographic level is important. Pre-

vious work was constrained both by the availability of reliable data to measure culture

and by identification. The novel dataset used in our worked and the allocation policy

implemented in the German context allow us to overcome these difficulties.

Finally, our work speaks to the literature that leverages the quasi-exogenous alloca-

tion of refugees within Germany to assess the effect of local characteristics on a wide

range of economic outcomes (Bahar et al., 2019; Battisti et al., 2016). More broadly,

we complement the growing literature on the causes and consequences of the post-2015

refugee inflow to Germany and Europe (Gehrsitz & Ungerer, 2017; Battisti et al., 2019;

Marten et al., 2019; Hangartner et al., 2019; Deole & Huang, 2020; Hilbig & Riaz, 2020;

Giavazzi et al., 2020; Busch et al., 2020).

The remainder of the paper is structured as follows. Section 2 describes the in-

stitutional background, and presents a simple conceptual framework linking the local

environment to refugees’ assimilation effort and success. Section 3 presents the data

and the construction of our proxy for cultural distance. Section 4 describes the empir-

ical strategy, presents the main results, and performs a number of robustness checks.

Section 5 examines the mechanisms. Section 6 concludes.

2 Background and Conceptual Framework

2.1 The Refugee Influx to Germany

Germany is one of the main destinations for refugees in Europe. Between 2015 and 2018

alone, a total of 1.6 million asylum applications were filed in Germany, amounting to

over 40% of all applications in the European Union during this time (Eurostat, 2019).

The surge in asylum applications followed the eruption of the civil war in Syria and the

growing threat of the so-called Islamic State in Iraq. Starting in 2011, an increasing

number of refugees fled to neighboring countries, moving westward to seek protection in

Europe. The movement of hundred thousands of refugees from Syria and Iraq through

Turkey and the Balkan Route, crossing Greece, Serbia, Croatia, or alternatively Hungary,

6

rippled into an even larger and more diverse movement of people, including asylum

seekers from Albania and Kosovo.

The number of asylum applications in Germany peaked in late 2015, following An-

gela Merkel’s highly contested decision in September of 2015 to admit refugees that

were stranded in Hungary (Figure 1). This decision was a deviation from the Dublin

Regulation, which assigns the responsibility of administering an asylum request to the

country of first-entry. However, the regulation was effectively (though not officially)

abandoned before September 2015 as registration and administrative capacities in Italy

and Greece ached under the immigration pressure and many refugees desired to continue

their journey and seek refuge in Northern Europe.

In an effort to curb the number of refugees, in March 2016, the European Union (with

Germany in a leading role) established a treaty with Turkey that encouraged stricter

controls by Turkish authorities at its Western shores. Under this agreement, Turkey

would take back refugees from Greece, and would resettle local refugees in the European

Union. This treaty, in combination with the closing of the Southern Hungarian border,

led to a steep decline in asylum application in Germany, which have remained relatively

low (at pre-2014 levels) since then.

Despite early warning signs, such as increasing numbers of refugees in Iraq and Syria’s

neighboring countries and growing refugee inflows across Europe, German authorities re-

mained ill-prepared for upcoming influx. The accommodation of hundreds of thousands

refugees within a few months proved to be a major challenge for German authorities.

One major tool for the distribution of refugees across States (Bundeslander) was the

so-called Konigsteiner Schlussel, which allocated refugees according to a State’s eco-

nomic capacity (tax revenues) and population. States themselves could then distribute

refugees within their districts, following independent but similar criteria. Focusing on

2016, Figure 2 shows that the local presence of refugees is indeed consistent with the

distribution that would have arisen under the the assignment through the Konigsteiner

Schlussel.

Refugees were often allocated according to the availability of housing at the local

level, taking into account basic demographic characteristics of the refugee (such as age,

gender, family status, and country of origin). However, for the most part, the pace of

refugee arrivals left no room for either one on one conversations with assignment officers

or in-depth analyses of refugees’ profiles. Within a short period of time, the available

accommodations were filled up and local authorities had to rely on alternative solutions,

such as vacant houses, empty hotels, old military barracks, schools, and improvised

7

container colonies and tents (BAMF, 2018).

Beyond the initial assignment to accommodations within states, refugees, in princi-

ple, had the ability to self-relocate under certain circumstances. Those who were still

in the asylum application process or who had already been rejected were not allowed to

move within the first three months of stay in Germany. Many of rejected asylum appli-

cants receive a special status, by which they are not officially refugees but whose stay

in the country is tolerated (Duldung). Until August 2016, accepted applicants as well

as persons with Duldung and pending applications that passed the three month mark

were allowed to move freely across Germany. Economic pull factors and large secondary

migration fueled the fear of parallel societies if refugees were to choose their place of

residence freely. Consequently, law makers passed the Integration Act in summer of

2016, restricting the free movement of refugees across states for the first three years in

an attempt to avoid “refugee ghettos” in big cities. Six out of sixteen states (mainly the

wealthiest and most densely populated states, such as Bavaria, Baden-Wurttemberg,

and North Rhine-Westphalia) tightened the law further, prohibiting refugees to move

out of the districts they were initially assigned to.

Although we exploit the residency obligation policy as a robustness check, its in-

troduction and the exogenous allocation by authorities across regions is not crucial for

our identification strategy, as we describe in more detail below. In contrast with most

others papers that focus on the German refugee setting, we are less concerned about

the random allocation, and more concerned about the consistent allocation of refugees

over time. Since we study the assimilation trajectories of refugees, we are less concerned

about initial placement (as this is a difference in levels) and focus on the consistency

of placement policies over time to track the speed of assimilation. We return to these

points below, when presenting the empirical test of our identifying assumptions.

2.2 Conceptual Framework

The literature in sociology (see Reitz, 2002, for a review and Phillimore, 2020, for an

application to the refugee context) and policy makers view “integration [as] a dynamic,

two-way process of mutual accommodation by all immigrants and residents”(Council

of the European Union 2004: 19). Yet, the distinction between assimilation effort and

success is, with very few exceptions, not made explicitly in economics. Typically, this

literature uses inter-marriage rates, naming patterns, and naturalization as measures

of social and political integration of immigrants (Abramitzky et al., 2020; Algan et al.,

8

2013; Bleakley & Chin, 2010; Meng & Gregory, 2005), and labor market outcomes (e.g.

wages or occupational income scores) as proxies for economic assimilation (Abramitzky

et al., 2014; Borjas, 1985; Eriksson, 2020). However, these outcomes, when considered in

isolation, reflect only one of the two sides of assimilation. Additionally, data limitation

have often forced researchers to focus on assimilation over the long run.

In the German refugee context, we can examine both determinants of integration,

testing how each varies with the level of “threat” in the area where refugees settle in

the short run. Building on the sociology literature mentioned above, we view successful

assimilation as the product of two forces: assimilation effort exerted by refugees and the

attitudes and behavior of locals.8 On the one hand, refugees choose how much – if any at

all – costly effort to exert in order to learn about and adopt local habits and norms. On

the other hand, locals incur a psychological cost when interacting with refugees. Such

cost is a decreasing function of (actual or perceived) cultural distance between locals and

refugees, which is in turn influenced by the effort exerted by the latter. Note that the

level of effort chosen by refugees may depend on locals’ openness (or, hostility), but locals

cannot directly interfere with such choice. For instance, locals cannot prevent refugees

from adopting a given cultural norm or from choosing a specific name for their offspring.

At the same time, effort alone is not enough to guarantee successful integration, which

instead requires also locals’ willingness to accept refugees in the majority group.

Mapping these insights to our context, in each month refugees can decide to exert

costly effort to become closer to local culture. With a certain probability, which depends

on assimilation effort accumulated until then, refugees successfully assimilate, and enter

the labor market, where they earn a wage higher than the value of their outside option.9

Yet, successful assimilation also depends on the local environment. While many local

factors might promote or hinder assimilation, we focus on the level of threat potentially

faced by refugees. In xenophobic and hostile societies, refugees need to exert more

effort to assimilate. That is, ceteris paribus, the same level of effort allows a refugee

to successfully assimilate earlier in a region with lower levels of hostility. However, and

crucially, the level of effort is not fixed, but is instead chosen by refugees in response to

local conditions.

When the cost of exerting effort is too high (relative to the probability of assimilating,

and the resulting benefits), refugees may decide not to exert any effort (Fouka et al.,

2021). In this case, forced assimilation backfires, and immigrants fail to assimilate

8For a more formal discussion of this framework see also Fouka et al. (2021).9For simplicity, we proxy for successful assimilation using labor market access and returns to successful assimilation

as wages. Assimilation can be more broadly defined as obtaining access to social and economic opportunities.

9

(Abdelgadir & Fouka, 2020; Fouka, 2020). However, as long as the level of effort is

strictly positive, refugees exert more effort in more hostile environments. Moreover, in

more threatening environments, refugees are likely to exert more effort earlier (compared

to refugees in more friendly regions), leading to non-linearities.10 Also, if refugees can

choose along which dimensions to exert effort, in a more threatening environment, they

may first invest in traits that are more easily observable. That is, consistent with a

signaling strategy, in a more hostile environment, refugees are going to converge faster

to local culture along more observable traits. Despite the higher level of effort, though,

refugees may end up assimilating later in areas where threat and discrimination are

higher.

Appendix Figure A1 captures graphically the main intuition of our conceptual frame-

work. The blue line represents refugees’ assimilation path, Ct, in a low threat environ-

ment, where the threshold level of assimilation is set to C by locals. In this setting,

refugees successfully assimilate at time t∗. In a region with a more hostile environment,

the threshold level set by locals increases from C to C′. Facing a higher threshold,

refugees adjust their effort, and their assimilation path (depicted in red in Figure A1) be-

comes steeper. Refugees trade off the cost of exerting effort and the gains from successful

assimilation. Despite the steeper assimilation curve (which captures higher assimilation

effort), refugees in a more hostile environment end up successfully assimilating later, at

time t′′∗ > t∗.

Our discussion also relates to a large literature that has shown how ethnic networks

may facilitate the economic integration of refugees, especially in the short run (Battisti

et al., 2016; Edin et al., 2003). At the same time, a larger ethnic network may reduce

incentives for refugees to exert effort, thereby retarding social (Gagliarducci & Tabellini,

2021; Lazear, 1999) and, possibly, economic (Borjas, 1985; Eriksson, 2020) assimilation.11

Overall, this section delivers a number of predictions that we can test empirically in

our data: i) effort should be higher, but successful assimilation lower, in more threat-

ening environments (as long as refugees choose to exert any effort at all); ii) in more

hostile environments, effort, but not successful assimilation, may display non-linearities,

as refugees decide to exert more effort early on; iii) larger ethnic networks should be

associated with lower effort, while their impact on successful assimilation is ambiguous.

10Such non-linearity is instead unlikely to be observed for successful assimilation.11Relating ethnic enclaves to our previous discussion, if networks grant access to an alternative set of social and economic

opportunities that require a lower C, then refugees may decide to reduce their assimilation effort. The effect on successfulassimilation, as proxied by labor market participation, is ambiguous and depends on the relative returns to a network jobversus regular job.

10

3 Data Sources and Measure of Cultural Distance

3.1 Data

Our analysis relies on two main survey datasets: the German Socio-Economic Panel

(SOEP) and the IAB-BAMF-SOEP Survey of Refugees.

The German Socio-Economic Panel. The German Socio-Economic Panel (SOEP)

is a large, nationally representative longitudinal study that surveys around 15,000 house-

holds and about 30,000 individuals every year since 1984, mostly in face to face inter-

views. The SOEP includes rich information on demographics, the socio-economic sta-

tus, and migration background of respondents. Also, and crucially for our purposes,

the SOEP reports the state, the region, and the district of residence of respondents.

This allows us to construct a measure of local culture that we can match to the answer

given by refugees at the same level of aggregation (district, region, and state). For more

information on sampling, fieldwork, data structure and content of the SOEP, we refer

to Goebel et al. (2019).

The refugee survey. We complement the SOEP with a second dataset, which allows

us to measure refugees’ preferences over time: the IAB-BAMF-SOEP Survey of Refugees.

This is a unique, novel dataset that contains detailed information and includes a large

number of refugees in Germany. The dataset is a longitudinal, representative survey of

refugees, asylum seekers, and their family members in Germany (Brucker et al., 2016).

The survey is conducted jointly by the Institute for Employment Research (IAB), the

Research Center of the Federal Office of Migrants and Refugees (BAMF FZ), and the

Socio-economic Panel (SOEP) at the German Institute for Economic Research (DIW

Berlin). The sampling frame of the survey is the Central Register of Foreigners in

Germany, where each foreign citizen is registered by her or his legal status. The target

population is composed of individuals arrived as asylum seekers in Germany between

January 1, 2013, and December 31, 2016, irrespective of their current legal status. The

total sample includes about 8,000 adult respondents (18 years and older), who were

surveyed up to three times between 2016 and 2018. More information on the sampling,

fieldwork, survey instruments and response behavior can be found in Kuhne et al. (2019).

The main questionnaire includes more than 400 questions regarding migration, em-

ployment and education history, socioeconomic and demographic characteristics, health

status, measures of social and political integration, as well as values and attitudes. This

11

dataset is complemented with a survey conducted at the household level that asks ques-

tions about housing, living conditions, and welfare benefits. Conveniently for us, both

refugee surveys are linked to the SOEP dataset. Moreover, the refugee survey is designed

to match as close as possible the questions in the SOEP, and the sample as well as the

interview process are similar between the two surveys. This important to ensure the

comparability of the two surveys – a key condition to study differences in values and

attitudes between refugees and locals.

Measuring the threat environment. An important part of our analysis below is

based on the presence of threat at the local level, which we measure in different ways.

First, we use the vote share for the Alternative for Germany (AfD) in 2017 federal elec-

tions. Because the AfD adopted a clear anti-refugee rhetoric only after 2015, we also

include the vote share for the far-right National Democratic Party (NPD) in the 2013

elections.12 The data are collected from the Federal Election Commissioner, and are

available at the district-level for both federal election cycles of 2013 and 2017 (Bun-

deswahlleiter, 2020). We then aggregate them at the NUTS-2 region level. Second, we

collect geo-located and date-stamped data on hate crimes against refugees from Bencek

& Strasheim (2016).13 We then calculate the first principal component of each measure

of threat in order to create a unique “threat” index.

We plot the threat index across NUTS-2 regions in Figure 4, both unconditional (left

panel) and conditional on state fixed effects (right panel). In Appendix Figure A2, we

also present the (unconditional) distribution of each threat dimension. Both the index

and its components display significant regional variation, and the individual dimensions

seem to be geographically correlated with each other. Overall, threat levels are most

pronounced in the former Eastern part of Germany – a pattern especially apparent for

the right-wing vote and hate crimes against refugees.

To validate the proxy for local threat, we exploit a specific question in the refugee

survey, which asks respondents whether they are concerned about xenophobia. In column

1 of Table A7, we report the coefficient on the threat index in a regression where the

dependent variable is a dummy equal to one if the refugee reports at least some concerns

about xenophobia. Even after controlling for survey year fixed effects, for months since

12Ideally, one would also measure support for the AfD at baseline, i.e. 2013. However, as noted in the text, the partyturned openly against immigration only after 2015, and its vote share in 2013 elections would thus not be indicative ofsupport for anti-immigrant stances. Reassuringly, as we discuss below, results are similar when using the NPD vote sharein 2013 and the AfD vote share in 2017, reducing concerns of endogeneity.

13One may be worried that hate crimes are endogenous to the presence of refugees. Below, we address this concern byfocusing on respondents interviewed after hate crimes are measured.

12

arrival as well as for both district and individual level characteristics, the threat index is

positively and strongly associated with worries about xenophobia among refugees. That

is, refugees assigned to regions with a higher threat index are significantly more likely

to report concerns about xenophobia, as compared to refugees assigned to areas with a

lower level of the threat index.14

Descriptive statistics: locals and refugees. The rich set of questions asked in

both the SOEP and the refugee survey allows us to investigate different dimensions of

what we loosely refer to as culture. One cluster of questions evolves around daily life

and leisure activities. Another block of questions deals with attitudes and personality

characteristics, locus of control, reciprocity, self-esteem, and risk aversion. The third,

and final, cluster of questions includes religion and faith, worries, satisfaction as well as

political interest. In addition to the battery of questions on culture, the survey includes

socio-economic characteristics, migration history, and other information about refugees’

conditions in the origin country. As discussed in more detail below when describing the

empirical strategy, we also rely on questions about the timing, the location, and other

conditions prevailing during the allocation process.

Table 1 presents the main variables of interest for our analysis, reporting the key

cultural dimensions separately for refugees (upper panel) and locals (lower panel) as

well as for high and low threat NUTS2-regions. Comparing preferences and values

expressed by refugees and locals, we see that risk aversion is higher among the former

than among the latter.15 This is consistent with the literature on risk-taking adjustment

after traumatic events (El Bialy et al., 2017; Ceriani & Verme, 2018). Refugees report

lower values of negative reciprocity, but higher values of positive reciprocity, relative to

natives. Interestingly, satisfaction with one’s health and housing conditions are virtually

the same across locals and refugees, while refugees engage in fewer leisure activities. As

expected, interest in politics is lower for refugees.

Examining preferences of refugees assigned to regions below and above median of

the threat index in our sample, we note that they are remarkably close to each other.

This suggests that refugees assigned to more or less hostile regions are similar, at least

along their observable (cultural) characteristics. Interestingly, a similar picture stands

out when focusing on locals’ preferences, in the lower panel of Table 1.

Next, Table 2 provides an overview on individual-level controls. In our sample,

14Since refugees are potentially lees likely to report the full extent of their fear of xenophobia in high threat environments,we take this as a lower bound estimate of the true effect of threat on refugees’ concerns.

15We show the detailed framing of the question in Table A1.

13

refugees’ average time spent in Germany is 29.5 months. About two in three respondents

in the refugee survey gained official asylum seeker status by the time of the interview,

with 20% pending decisions and 16% rejected applicants (not reported for brevity).

Syria, Afghanistan, and Iraq make up for 78% of the regions of origins of refugees in

our sample. Finally, turning to socio-demographic characteristics, the refugee sample

is younger (34 years versus 51 years on average), more likely to be male (only 38% of

refugee respondents are female), and less educated.

Additional datasets. We complement the datasets described above with additional

data sources. First, we obtain total population and the number of refugees at the time

of the survey (end of year) at the district level from the German Federal Statistical

Office. Second, we retrieve data on regional unemployment rates for each survey year

and district and the employment rates and median wage of immigrants (NUTS2-, region-

of-origin- and time-specific) from the statistics department of the Federal Employment

Agency (Statistik der Bundesagentur fur Arbeit, BA, 2020). Table 3 presents summary

statistics for these additional variables measured either at the district or at the NUTS2-

level.

3.2 Measuring assimilation effort and success

The context (and the data) considered in our work offers the possibility of measuring

both margins of assimilation – effort and success – in the short run (between one to

five years). Specifically, we measure successful assimilation as the labor market status

reported by the refugee at the time of the interview. As discussed above, labor market

integration clearly depends on locals’ willingness to cooperate and engage with refugees.

Although we lack a direct proxy for social assimilation, such as intermarriage (Gordon,

1964), we view labor market status as a good measure not only of economic but also of

social integration.

To measure assimilation effort, we build on the existing literature (Alesina et al.,

2017; Bertrand & Kamenica, 2018; Desmet et al., 2017; Desmet & Wacziarg, 2018), and

exploit high frequency attitudinal data from the refugee survey. We construct a measure

of cultural proximity between the refugee and locals living in the same NUTS2 region.

Exploiting the large sample size of the Socio-Economic Panel, we define a measure

of refugees’ similarity to local, rather than national, culture by taking the average of

preferences expressed by local residents before the influx of refugees. This measure

14

allows us to examine whether, over time, refugees express preferences and norms that

are closer to those reported by locals at baseline. Since cultural, political, and social

values reported in the survey by refugees do not depend on the actions of natives, they

are a good proxy for the effort side of assimilation. As already discussed above, the

decision to exert effort may be influenced by the local environment, but, once such

decision is taken, locals cannot alter it.

Since we are interested in convergence to local culture, we need sufficient variation

of locals’ cultural preferences within a given geographic unit. Thus, we face a trade-off

between granularity and representativeness. Although we can observe respondents’ lo-

cation at the district level, some districts contain as little as 20 non-refugee respondents.

For this reason, we prefer to use a higher aggregation level: the NUTS2 region.16 In a

robustness exercise, however, we replicate the analysis “zooming in” down to the district

level. Focusing on positive and negative reciprocity (see Table A1 for the exact wording

of the questions and the range of possible responses), Figure 3 illustrates the importance

of within country (and even within state) cultural heterogeneity prevailing among locals

in 2010 (i.e. well before the influx of refugees) across NUTS2 regions.

There is an array of different statistical measures that can be used to capture distance,

entropy, or divergence (Cha, 2007). Desmet & Wacziarg (2018) define cultural distance

between two groups as the share of total heterogeneity in responses to questions in

the General Social Survey (GSS) that is not attributable to within-group heterogeneity.

This measure belongs to the group of “overlap measures”.17 Another commonly used

measure is the Euclidean Distance, which belongs to the group of geometric distance

measures. These measures are derivatives of the Minkowski norms, which is written as

Dmink(X, Y ) = p√∑n

i=1 |xi − yi|p, with X and Y as two independent probability density

functions. The Euclidean Distance is part of the Minkowski family (with p = 2), and

one of the most widely applied measures of cultural distance (Rapoport et al., 2020;

Bertrand & Kamenica, 2018; Alesina et al., 2017). Intuitively, it captures the shortest,

unweighted distance between two points in the cultural space.

Following the literature cited above, we use the Euclidean distance to capture the

cultural proximity between a single refugee and local (non-refugee) residents in the same

NUTS2 region. For each question (see Table A1 for the list of all questions included in

16Germany has 38 NUTS2 regions, which gives us a sufficient number of observation per region to reduce measurementerror and at the same time highlight the relevance of local culture.

17Desmet & Wacziarg (2018) define eleven identity cleavages (including age, education, ethnicity and other demograph-ics) and create a cultural distance measure that predicts how two individuals from various groups picked at random wouldgive the same response to an answer.

15

the index with the corresponding definition), we first calculate the pairwise differences

between the refugee and all locals, xi−yi. Then, we square those differences and take the

mean. Finally, we calculate the square-root of this term so as to obtain the Euclidean

distance between the individual refugee and all individuals living in the same NUTS2

region for a specific question DEucl(X, Y ) = 2√∑n

i=1(refugeei − locali)2. We then take

the mean Euclidean distance over all questions, and invert this term to get a cultural

similarity measure, which we summarize in Table 1.18. In an additional exercise below,

we also consider an equivalent measure that captures the cultural distance between

residents of the NUTS2 region and the rest of Germany (as shown in Figure A5) at

baseline. We calculate the Euclidean distance using the pairwise differences between the

average local response and average national response to a specific question.

To isolate refugees’ convergence to local culture, we fix responses of local residents

at baseline. While locals’ preferences may change in response to refugee inflows, making

our baseline measure less accurate, we want to prevent our proxy for cultural conver-

gence from being possibly influenced by locals “moving closer” to refugees, as shown

by Giuliano & Tabellini (2020) for the US context. Whenever available, we take the

locals’ responses to a specific question in the year before the large influx of refugee in

2014. When this is not possible, we use the closest observation year possible in case

the question was not asked in 2013. With the exception of religion, all questions were

asked before 2013 (either in 2012 or in 2010). This guarantees that, indeed, our index

of cultural proximity is constructed using pre-determined preferences of locals.19 The

cultural dimensions used in our analysis naturally arise from the overlapping questions

in the refugee and general population survey. We make no a-priori assumption about

which questions best represent local culture. Instead, we incorporate all questions that

are available for both refugees and locals into our index. Below, however, we also disag-

gregate the index into each of its sub-components to explore heterogeneity.

4 Empirical Strategy and Baseline Results

In this section, we first describe the empirical strategy (Section 4.1). Next, we present

baseline results for cultural convergence and economic assimilation of refugees (Sec-

tion 4.2). Finally, we discuss and provide evidence to corroborate the validity of the

18We illustrate the average unconditional cultural similarity between refugees and locals of the same NUTS2 region inFigure A3

19Results are robust to omitting religion, which is measured for the first time in 2016.

16

identification strategy, addressing concerns about non-random assignment of refugees

and ex-post selection of refugees across regions (Section 4.3).

4.1 Empirical Specification

To study how refugees’ preferences and their employment status change with each extra

month spent in a German region, we estimate:

Yidt = α + β1MSAit + β2MSA2it + β3X

′it + β4Y

′dt + γd + γt +Qit + εidt (1)

where Y is either cultural similarity (relative to the local population) or employment of

refugee i in district d and survey year t.20 The main regressor of interest is months since

arrival of the refugee, MSA. In order to capture potential non-linearity in assimilation,

we also include the square of months since arrival.

We further control for: i) baseline district level variables (unemployment rate, pop-

ulation density, and share of asylum seekers) interacted with time dummies, Y ′dt; ii)

individual characteristics (gender, age, marital status, work experience, education and

country of origin), X ′it; and, iii) refugee specific time-varying dummy variables, Qit, to

account for compositional changes in the questionnaire and refugees’ responses (or miss-

ing values). This last set of controls is particularly important when focusing on cultural

similarity, because we want to ensure an adequate comparison group. That is, we want

to compare refugees that answered the same set of attitudinal questions over time. Stan-

dard errors are clustered at the person level to account for the fact that some refugees

are surveyed repeatedly, following the sampling-based clustering approach proposed by

Abadie et al. (2017).21

In our preferred specification, we include district and interview year fixed effects in

order to absorb, respectively, any district-specific (time invariant) characteristics and

trends common to all refugees over the sample period. Moreover, we use the region

of assignment – rather than the region of residence – as the location of treatment,

thereby implementing an ITT approach. This is possible because the refugee survey

distinguishes between the assignment and the residence location (about 25% of refugees

in our sample have moved outside of their assigned NUTS2 region). In the second part

of the empirical analysis, we more explicitly test the “threat hypothesis”, by augmenting

20When examining refugees’ employment, we restrict attention to survey respondents in the age range 18-64.21Reassuringly, however, results are robust to clustering standard errors at either the district or the NUTS-2 level to

account for potential spatial correlation in the error term.

17

our (most stringent) specification with interactions between months since arrivals – both

linear and squared – and the threat index described above.

The key identifying assumption behind our empirical strategy is that the allocation

of refugees to German regions does not change over time. For instance, one may be

worried that, over time, officials become better able to “match” refugees to the region of

assignment on the basis of their cultural similarity. Alternatively, it is possible that, due

to the rising number of asylum seekers, refugees arriving later may be assigned to areas

with worse economic outcomes, where their integration is harder. In either scenario, our

estimates would be biased due to ex-ante sorting of refugees across regions.

A second set of threats to our empirical strategy is related to possible ex-post migra-

tion of either locals or refugees. Using an IIT approach addresses the possible migration

decision of refugees. However, it does not deal with the fact that locals with varying

degree of openness may selectively move away from regions that receive more refugees.

If this pattern were to change over time, our results may be driven by selection. Even

if we construct the index of cultural similarity between refugees and locals using the

culture prevailing in the region at baseline, such population changes may nonetheless

change both incentives for refugees to exert effort and their eventual assimilation.

After presenting our main results in the next section, we return to these issues in

Section 4.3, and provide several pieces of evidence that assuage these, and additional,

concerns.

4.2 Baseline Results

Assimilation Effort

In Figure 5, we present the relationship between local Euclidean cultural similarity and

months since arrival in the raw data, without any control. A clear positive trend, with

similarity increasing over time, emerges. To go beyond such correlation, we report more

formal regression results in 4, multiplying coefficients by 100 to ease the interpretation.

In column 1, we report the correlation between months since arrival and our cultural sim-

ilarity index (CSI), absorbing survey-question composition fixed effects and controlling

for months since arrival (MSA) squared. The coefficient on MSA is positive, quanti-

tatively large, and statistically significant, confirming the pattern already displayed in

Figure 5.

Next, in columns 2 to 7, we gradually introduce a more stringent set of controls. In

column 2, we begin by adding survey year fixed effects, while in columns 3 and 4 we

18

include individual-level and district-level time varying-controls. Columns 5 and 6 further

include state and NUTS-2 region fixed effects. Finally, column 7 controls for district

fixed effects, thereby comparing local convergence between refugees assigned to the same

district in different months. For the remainder of the paper, we consider column 7 as

our baseline specification22. We replicate our findings using an alternative statistical

measure for cultural similarity – the Canberra index23 – in Table B1 and the results

hold.

In all cases the coefficient on MSA remains positive and statistically significant,

indicating that refugees gradually converge to local culture as they spend more time

in a region. Moreover, we find it reassuring that the point estimate remains virtually

unchanged when including additional controls, indicating that the allocation process is

unlikely to be influenced by factors that may vary over time and influence the assimilation

trajectories of refugees. The negative, albeit small, coefficient on the MSA squared term

is consistent with the speed of assimilation falling over time.

One possible way to gauge the magnitudes of our estimates is to ask when the

average cultural similarity between a refugee and a local would equal the average cultural

similarity between locals themselves. We thus calculate the Euclidean cultural similarity

index between all locals in the same region using the pairwise difference between locals

(rather than that between refugees and locals). As shown in the right panel of Table 1,

the average cultural similarity between refugees and locals lies at -1.74 – as expected,

lower than the average distance between locals (-1.37). According to our preferred

specification (column 7 in Table 4), after one year, refugees are able to close 5% of

this gap. Assuming a linearity – an imperfect, but not unreasonable assumption given

the very small coefficient on the MSA squared term – refugees’ average culture fully

converges to average culture of locals within 20 years.24

Our analysis thus far has focused on average convergence. In Table A3, we present

results separately for the various items included in the cultural similarity index. The

confidence intervals of our coefficients are adjusted for multiple hypothesis testing, fol-

lowing Clarke et al. (2019); Romano & Wolf (2016, 2005a,b). We find convergence for

22In Appendix Table A2 we also report coefficients on individual and district level controls23The Canberra index is one of the many derivatives in the Minkowski family (to which the Euclidean distance belongs)

of statistical dissimilarity. It writes as DCa =∑d

i=1|Pi−Qi|Pi+Qi

, with Pi and Qi representing two probability density

functions. In comparison to the Euclidean distance, the Canberra distance decreases the weight on outliers. In otherwords, if refugees converge to locals only along one cultural dimension, this would be captured in the Euclidean indexand would be discounted in the Canberra index.

24Since we are not able to follow refugees for such a long time, we cannot rule out stronger non-linearities after a fewyears. However, the magnitude of our results is quantitatively large, and we believe that can offer us some insights aboutthe short- to medium-run dynamics of cultural assimilation.

19

positive reciprocity, social inclusion, leisure activities, life satisfaction and interest in

politics. Since Table A3 focuses on convergence relative to locals’ preferences, however,

it does not tell us in what direction refugees’ preferences change over time. We thus

complement this analysis by plotting the change in refugees’ preferences by arrival co-

hort, after partialling out individual controls, district level time-varying characteristics,

and district fixed effects. We present these estimates in Figure A4. For some cultural

traits, we find consistent patterns, where more time spent in Germany also translates

into, for instance, more similar kinds of leisure activities, interest in politics and feelings

about social inclusion. These trends are to be expected as refugees become more inte-

grated into “local life” of the region. However, some dimensions of convergence – such

as reciprocity – do not display consistent patterns. Refugees converge to the regional

levels of reciprocity but they do not generally become more or less reciprocal with time

spent in Germany.25

Assimilation Success

In Table 5, we turn to our key proxies for successful (economic) assimilation.26 In the

top panel, the dependent variable is a dummy equal to one if the refugee is employed at

the time of the interview. In the bottom panel, we instead consider the (log of) wage,

restricting attention to employed refugees. The structure of the table mirrors that of

Table 4. Focusing on the most stringent specification (reported in column 7), we note

that both employment and wages increase as refugees spend more time in the region

of assignment. As for cultural convergence, the point estimate on the quadratic MSA

term is negative, although quantitatively very small (and not statistically significant for

wages).

In both cases, the coefficient on MSA is statistically significant and quantitatively

large. According to our preferred specification, one extra year in the region of assignment

raises the probability of employment by 9.6 percentage points (or, 50% relative to the

sample mean) and wages by 8%. Although cultural convergence may be responsible

for the increase in employment and wages of refugees, the opposite relationship (with

economic assimilation increasing incentives to adopt local culture) may be at play. In

our analysis, we do not aim at identifying the causal effect of “effort” on assimilation

success. Rather, we are interested in separately estimating the reduced form relationship

25If reciprocity varies substantially across regions (as indicated in Figure 3), average cohort effects might hide thisheterogeneity.

26The analysis in Table 5 is restricted to individuals in the age range 18-64.

20

between both cultural similarity and economic integration on the one hand and months

spent in Germany on the other.

In Section 5, we examine how these relationships vary with the local environment,

focusing specifically on threat. Before doing so, in the next section, we present a number

of checks to corroborate the robustness of the results presented thus far.

4.3 Assessing the Validity of the Identifying Assumptions

In this section, we perform several robustness checks to corroborate the validity of our

identification strategies, reporting all exhibits in Appendix B..

Selection on the side of authorities

While the refugees’ allocation process in Germany followed specific rules, some degree of

discretion may have remained for placement officers. For example, one may be worried

that state officials tried to match refugees with certain characteristics to specific districts,

and that this pattern changed over time. If such (change in) assignment were based on

observable characteristics, our analysis would deal with that by including individual

controls in the regressions. However, if refugees’ assignment changed over time along

non-observable dimensions, with state officials becoming better able to match culturally

individual refugees to German regions, our estimates may be biased.

While we cannot directly test this possibility, we compare the (observable) charac-

teristics of refugees, including baseline cultural similarity, arriving at different times and

assigned to different regions. Results for this exercise are reported in Table B2. In

Panel A, we compare refugees’ characteristics between NUTS-2 regions with baseline

unemployment above and below the sample median, while in Panels B and C we split

the sample by rural-urban status and threat index, respectively.27 In columns 1 to 4, we

interact the year of arrival with the following pre-entry characteristics: gender, age, work

experience at entry, and origin. We use as reference year 2015, which is the year with the

largest number of arrivals in our sample. Due to the limited number of observations, and

so as to have a balanced sample, we pool together individuals arrived in 2013 and 2014.

Reassuringly, coefficients are always statistically insignificant and quantitatively small.

Moreover, there is no evident pattern across regions for individual level characteristics.

27We define urban-rural status following the Federal Institute for Research on Building, Urban Affairs, and SpatialDevelopment.

21

As an additional check, in column 5, we conduct a similar exercise for cultural simi-

larity. Since we lack refugees’ attitudes before they enter Germany, we use their answers

in the first available survey. The first survey took place in 2016, and covers refugees

that arrived between 2013 and 2016; because of this, and to have a sufficient sample

size, we can only compare refugees that arrived in 2016 to those that arrived in 2015,

which leaves with only 382 observations. The small number of observations suggests

that one should interpret this result with caution. However, and reassuringly, also in

this case, we observe no statistically significant and no quantitatively large difference

between refugees arrived in different years.

Sorting on the side of refugees

An additional threat to identification could be ex-post self sorting of refugees. In prin-

ciple, even after refugees have been quasi randomly assigned to districts with certain

cultural or economic traits, they could then - over time - move to places that better

match their preferences or offer them better labor market opportunities. We address

this concern in three ways. First and as mentioned above, we use the first assigned

district, not the district of residence, and measure the cultural similarity to locals in

that region. This follows the logic of an Intent to Treat and would bias our results

towards a null finding, assuming that refugees would move to places that are culturally

more similar to them, to regions that offer better job prospects or regions that are less

xenophobic.

Second, we look for patterns in cultural and economic selection into internal migra-

tion. We compare the cultural similarity and employment outcomes of refugees to the

region of residency, rather than the region of assignment and compare those who stayed

to those who moved (about 75% of refugees stay in the region of assignment). We run

our baseline regression - now without the ITT logic - but include a dummy for whether

the refugee is a stayer (e.g. did not move out of his NUTS2 region of assignment) or a

mover. We illustrate the coefficient for the mover dummy in Figure 6. We present the

cultural similarity index, all its sub-dimensions and employment outcomes and do not

find any evidence for economic or cultural selection on the side of refugees - at least in

the short-run.

Additionally, we exploit the residency obligation introduced in the summer of 2016.

As mentioned in section 2, some refugees fell under the the Integration Act and were not

allowed to move freely across Germany. We have self-reported information on whether

22

refugees fall under this residency obligation and can - for a subset of refugees - create a

more objective measure of residency obligation that takes into account characteristics of

the refugee (marital status, month of arrival etc.) to predict whether the refugee must

have fallen under the residency obligation. In Table B3, we test whether our findings

vary by the mobility status of refugees. Column 1 replicates our baseline specification

(Table 4, column 7). Next, in column 2, we turn to refugees who fall under the resi-

dency obligation. Reassuringly, even in this case the coefficient is positive and, despite

the substantial drop in the sample size, remains marginally statistically significant. Im-

portantly, coefficients in columns 1 and 2 are not statistically different from each other.

In columns 3 and 4, we consider separately stayer and movers: also in this case, results

remain positive, quantitatively large, and statistically significant. Although the coef-

ficient for movers is slightly larger than that for stayers, the two are not statistically

distinguishable from each other.

Lastly and similar to the plausibility check for sorting into threat regions by place-

ment officers, we also make sure that refugees did not differentially select out of threat

regions. Again, we take our main measures of threat regions and check whether the

likelihood to move out of this region increases differentially in time. We interact our

main variable of interest MSA with our overall threat PC index, as well as hate-crimes

committed against refugees to see whether out-migration patterns differ in threat re-

gions. We present out results in Table B4, and find no differential out-migration when

we include the set of controls and fixed effects used in our baseline regression.

One remaining concern - unrelated to sorting on the side of authorities or selection

on the side of refugees - could be that that the probability of survey attrition increases

with assimilation effort, success or the threat environment. Similarly, refugees could

choose to not answer certain types of questions in the threat environment. While we

address the latter in our baseline regression by including a vector of dummies for the set

of questions that were answered by the refugee, we still test both of these possibilities

explicitly in Appendix Tables B5, B6. We first check whether refugees with different

assimilation effort are more or less likely to appear in the next wave and we check for

our three main threat measures whether specific types of questions were less likely to be

answered. We find no evidence that attrition or non-response is driving our results.

23

5 Mechanisms

In this section, we analyze the potential mechanisms behind the results presented above.

First, we document that refugees assigned to regions with higher levels of threat con-

verge faster, but are not more likely to be employed or earn higher wages (Section 5.1).

Second, we examine the role that ethnic networks play for both cultural convergence

and economic integration (Section 5.2).

5.1 Threat Environment

Incentives to assimilate are shaped not only by the characteristics of the refugee com-

munity, but crucially depend on natives’ attitudes and openness towards the foreign

born. As mentioned in our introductory section,a more open environment might make

it easier for refugees to converge culturally and economically as this facilitates social

and economic interactions and a lack of openness by the host community may inhibit

assimilation or even cause a backlash, with refugees being more likely to preserve their

own cultural norms in the presence of hostile attitudes of locals (Abdelgadir & Fouka,

2020; Bisin & Verdier, 2001; Fouka, 2020). The literature on the “threat hypothesis ”

emphasizes that increased hostility can even encourage assimilation as refugees may need

to signal their level of assimilation to gain access to the local community and related

economic and social opportunities (Fouka, 2019; Saavedra, 2018; Fouka et al., 2021; Bisin

& Tura, 2019). In our conceptual framework, we allow for both of these dimensions to

come to play and have added the important distinction between assimilation effort (as

an input factor) and assimilation success (as a equilibrium outcome). We reiterate the

predictions of our conceptual framework. Threat environments will increase assimilation

effort but will decrease assimilation success.

Assimilation Effort

We present the results for the effect of threat on assimilation effort in Table 6. In column

1, we interact MSA with the PCA index for the threat environment. When we focus on

the linear specification (Panel A), the interaction between MSA and the index is positive

and not statistically significant. However, consistent with the threat hypothesis, the

coefficient becomes statistically significant when we turn to the quadratic specification

(Panel B). That is, in places with higher threat refugees’ cultural convergence happens

more quickly or in other words, refugees exert more assimilation early on. Deconstructing

24

our index into its four sub-components, we find similar patterns.28. Interestingly, a more

open local environment decreases assimilation effort of refugees linearly. Overall, we

observe a more speedy assimilation effort made by refugees in threat environment and

presumably lower incentives to culturally assimilate in regions that accept diversity and

therefore may allow refugees to access social and economic opportunities without having

to signal assimilation. We take this as strong evidence for our empirical proposition one.

We visualize our results in Figure 7. Based on our baseline specification in column

7 of Table 4, we predict assimilation effort trajectories of refugees over time. On the

y-axis we plot the level of cultural similarity (Euclidean index) and on the x-axis, we

plot months since arrival. We add the effect of the interaction terms between MSA and

MSAS with the threat environment, specifically vote for the AfD in 2017, vote for the

NPD in 2013, hate-crimes against refugees and openness of locals to the baseline effect

to simulate the differential trajectories over time. Similar, to our model visualization we

find in the simulation that threat environments increase the speed of assimilation effort.

Assimilation Success

An additional implication of this conceptual framework is that refugees’ labor market

outcomes – a proxy for successful assimilation – should be lower in places with more

discrimination and with a less welcoming environment. We present our findings in Table

7. As the dependent variable we use refugees’ self-reported employment status at the

time of the interview. Two results stand out. First and in accordance with proposition

2 of our conceptual framework, the relationship between assimilation success and the

interaction terms is flipped: employment probability is lower where the local environment

is more hostile.29 Second, the relationship seems to be linear; this is to be expected,

since the effects of discrimination on equilibrium outcomes like employment are unlikely

to be non-linear or to jump at some specific threshold. These results are, in our view,

strongly consistent with the threat hypothesis of assimilation.

Robustness Checks and Additional Tests

In addition to the battery of plausibility checks for our identifying assumptions, in this

section we will provide further robustness checks and investigate corollary implications of

28In a robustness check, we restrict the refugee sample to those who have arrived after the hate-crimes were measuredand find qualitatively the same results

29The interaction with hate crimes keeps the sign, however, loses statistical significance when restricting the sample toindividuals having arrived in 2016 or later (due to possible concerns about endogeneity mentioned previously.)

25

our conceptual framework. One concern might be that the threat environment is actually

capturing economic disadvantages that impact the cultural and economic assimilation

trajectories of refugees. While we control for district-level time-varying unemployment

rate, there may be the concern that the interaction term between MSA and the threat

environment captures some underlying economic conditions that are correlated with

right-wing voting or hate crimes against refugees. In order to rule out this, we include the

interaction between MSA and MSAS with the local unemployment rate to our estimating

equation. We present our results in Table B7 and find that including the interaction

with local unemployment does not alter our main findings.

Next, we test an assumption we have made in our conceptual framework. Specifically,

for our empirical predictions to hold, we assume that the required level of assimilation

is higher in threat environments. This means that at a given level of assimilation effort

(in our case cultural similarity), refugees are less likely to successfully assimilate in

threat environments versus neutral environments. We could therefore include a triple

interaction between MSA, the threat environment and the cultural similarity of a refugee

to gauge this effect. However, cultural similarity at time t might be endogenous to having

a job, since assimilation trajectories might change once refugees access the labor market.

We can remedy this concern for a subset of refugees that was interviewed multiple times

between 2015 and 2018. For these refugees we can include a lagged variable of cultural

similarity to check whether locals in threat environments indeed require higher levels

of assimilation. We expect that – given a level of cultural similarity and months since

arrival – the likelihood of being employed in a threat environment will be lower. We

estimate the triple interaction and report the results in columns 3 and 4 of Table A4. We

confirm that given a level of cultural similarity, refugees are less likely to be employed

in threat environments than in neutral environments.

Another implication of our conceptual framework is that refugees should decrease

their assimilation effort in threat environments more than in neutral environments, once

they have successfully assimilated. If indeed similarity in stated preferences serves as a

one-sided signaling tool to access economic and social opportunities especially in threat

environments, then we should observe a discontinuity in the trajectory of cultural assim-

ilation for employed refugees. We therefore expect the triple interaction between MSA,

threat and employment status of the refugee to be negative. We show the results in

Table A4, where we add the triple interaction for the lagged employment status of the

refugee to reduce concerns about reverse causality. In line with our prior, we find a neg-

ative effect of employment on subsequent assimilation effort in threat environments as

26

compared to more neutral environments. This could be because refugees in employment

do not have to exert as much effort since they have overcome a substantial hurdle but

it could also be that locals in threat environments do not engage with refugees in a way

that encourages cultural exchange on the job and therefore reduces cultural convergence.

Both of these interpretations are in line with the threat hypothesis of assimilation, ei-

ther conforming the signaling dimension or the non-cooperative/one-sided dimension of

cultural assimilation.

In order to exploit the heterogeneity in exposure to threat at the individual and

not at the regional level, we make use of another feature in the German allocation

policy: allocation to refugee shelters versus decentralized housing. We hypothesize that

refugees that live in refugee shelters must feel more vulnerable to outside threats as they

are more easily identifiable as the outside group. Living in a shelter should therefore

differentially increase fear if locals are more xenophobic. Given a certain set of controls,

the allocation to different types of shelters is plausibly exogenous and mainly depends on

availability. We run a triple interaction between MSA, threat and type of accommodation

and show in Table A5 consistently positive coefficients for both the linear and polynomial

specifications.

Lastly, we move to the distributional features of local culture. Assimilation pressure

may not only come from immediate threat or anti-immigration sentiment but can also

arise in the presence of a strong and distinct local culture. We conjecture that refugees

should also increase their assimilation effort in regions, that have a distinct local culture

and regions that are also internally homogeneous.

Our main assimilation effort outcome captures regional – and not national – culture

and preferences. However, based on the evidence presented thus far, we cannot rule

out the possibility that refugees may be converging to national culture, and we may be

incorrectly attributing this to local factors instead.30 In Figure A5, we show the cultural

similarity index between residents of the NUTS2 region and the rest of Germany, using

the whole set of questions listed in Table A1. We find that local culture can vary

substantially from the average national culture.

First, we begin with the hypothesis that assimilation effort increases in regions where

the local cultural is more distinct from national culture. To test this hypothesis, we

calculate the Euclidean cultural distance between the region and the whole country.31

30It is of course possible that refugees simultaneously converge to German national culture, while also absorbing thelocal preferences of the region where they happen to reside, at least along some dimensions.

31In particular, we calculate the standardized pairwise differences between the average local and the average national,which we square and add over all questions before taking the square-root.

27

We then interact it with months since arrival (MSA), and report results in Table A6,

where we augment our baseline specification (Table 4, column 7) with the interaction

between MSA and national-local cultural distance.32 In column 2, the coefficient on

MSA captures the convergence to the part of local culture that is shared by the average

German. More importantly for us, the coefficient on the interaction term captures the

convergence to the ”residual” culture, i.e. the culture that is specific to that region and

distinct from the views of the average German. As it appears, both the main effect and

the interaction term are positive and statistically significant, which we take as additional

evidence for assimilation effort resulting from assimilation pressure.

However, a distinct culture may still allow for diverse opinions within its ranks. We

therefore also examine the internal homogeneity of regional culture. We hypothesize that

deviations from the average local culture are ”punished” more – or are more obvious

– in regions, where locals have very similar preferences and norms. We expect that

assimilation effort increases more in homogeneous cultural environments, acknowledging

that internal homogeneity could also facilitate spontaneous cultural learning. Column 3

of Table A6 presents results consistent with this hypothesis.

5.2 Ethnic Networks

An extensive literature has analyzed the effects of ethnic networks on the process of

cultural and economic assimilation (Borjas, 1985; Lazear, 1999). On the one hand, a

larger ethnic enclave may help immigrants and refugees find a job, thereby promoting

their integration (Battisti et al., 2016; Edin et al., 2003). On the other, it might lower

incentives to exert effort to become familiar with and adopt local social norms and lan-

guage, in turn slowing down cultural (and potentially economic) assimilation (Eriksson,

2020).

To analyze if and how these forces operate in our context, we augment our preferred

specification (Table 4 and 5, column 7) by interacting MSA with the size of the ethnic

enclave and the degree of economic integration of its members. To reduce endogeneity

concerns, both mediators are measured at baseline. Results are presented in Table

8. In columns 1 to 3, we interact MSA with share of refugees relative to the NUTS2

population, at the end of the year prior to the survey year. In column 4 to 6, we consider

the employment share of immigrants from the same origin region.33 We present three

32We also include the interaction with the squared MSA (not shown in the table). For completeness, in column 1 ofTable A6 we replicate the baseline, non-interacted, specification.

33We classify refugees into five origin regions: MENA, Afghanistan, former USSR, West Balkans and Sub-Saharan

28

outcomes – cultural similarity, employment, and (log of) wages – reporting the linear

and the quadratic specifications in the top and bottom panels respectively.

Starting from cultural similarity (column 1), the coefficient on the interaction term

between MSA and the share of refugees in the region is negative and statistically sig-

nificant, both in the linear and in the quadratic specification. In line with previous

work (Advani & Reich, 2015; Eriksson, 2020), this indicates that individuals assigned

to regions with a larger ethnic enclave converge more slowly to local culture.34 The

negative effect of networks on cultural convergence, however, disappears when focusing

on the average employment of immigrants in the region (column 4). One possible ex-

planation is that the presence of more successful peers acts as a “role model” for newly

arrived refugees, who are in turn induced to exert more effort. This positive effect may

partly counterbalance the direct, negative effect of networks. Other interpretations are,

of course, possible, and we view this evidence as mostly descriptive.

In columns 2 and 3, we turn to employment and wages, respectively. In this case,

results are unstable, and no clear pattern seems to emerge. Even though the interac-

tion between MSA and the refugee share is negative and statistically significant in the

quadratic specification, coefficients are very noisy and positive in all other cases. These

null results may mask opposing effects of networks: on the one hand, ethnic enclaves may

increase employment opportunities within the refugee community or help refugees find a

job upon arrival; on the other, they may reduce skill accumulation and (as noted before)

lower cultural convergence, making it harder for refugees to integrate economically.35

The coefficient on the interaction between MSA and the mediator remains imprecisely

estimated also when considering the employment of immigrants in the region (columns

5 and 6). One may expect refugees to do better in areas with higher immigrants’

employment (at baseline), as this might indicate more opportunities for outsiders. Yet,

this force may be offset by the fact that in regions where immigrants are more likely to

be employed, refugees may face higher labor market competition.

Africa.34We obtain similar results (not reported for brevity) when considering as mediator the share of immigrants from the

same origin region.35A larger ethnic enclave may also increase locals’ discrimination, in turn further lower employment opportunities for

refugees.

29

6 Conclusion

In this paper, we examine the role of threat and locals’ hostility in determining the

assimilation trajectory of refugees. We provide a simple conceptual framework that

structures our intuition behind the impact of threat on assimilation and provides several

testable empirical predictions. We distinguish – both conceptually and empirically –

between the effect of threat on assimilation effort and success in the short-run (first 2

to 5 years), exploiting the quasi-random allocation of refugees to German regions to

establish a causal link.

We show that the ”cultural gap” between refugees and locals decreases by 5% every

year and that refugees increase their likelihood of being employed by 50% every year and

provide evidence that refugees increase their assimilation effort in threat environments,

especially early on (in the first 12 to 24 months after arrival) but are less successful at

assimilating. We also document that refugees exert more effort in places whose culture is

more “distinct” from the national average and that are internally more homogeneous. We

also find that larger networks unambiguously decrease assimilation effort and success, but

also that the quality of the network can have a positive effect on successful assimilation.

Our results are robust to an array of empirical exercises that aim to check the plausibility

of our identifying assumptions, particularly with respect to ex-ante selection and ex-post

sorting.

Our paper calls into question the effectiveness of pressure as a tool for integration.

While refugees may exert more effort in signaling allegiance to the local values and

norms in threat environments, this effort will eventually not translate into successful

assimilation as locals refuse to engage with what they perceive as an outside group.

With this analysis, we emphasize the role of locals in creating an environment where

refugees and immigrants more broadly can succeed and that integration is indeed a

two-sided process. Additionally, this paper calls into question the use of convergence in

preferences and norms as a measure of successful integration, if it is indeed motivated

by fear of social exclusion, discrimination or even violence.

We conclude by noting that, unfortunately, the number of forcibly displaced indi-

viduals is expected to increase dramatically in the years to come, and within country

diversity is likely to rise. Thus, understanding if and how cultural convergence takes

place will become increasingly important not only for economists and political scientists,

but also for policy-makers.

30

References

Abadie, A., Athey, S., Imbens, G. W., & Wooldridge, J. (2017). When should you

adjust standard errors for clustering? Technical report, National Bureau of Economic

Research.

Abdelgadir, A. & Fouka, V. (2020). Political secularism and muslim integration in the

west: Assessing the effects of the french headscarf ban. American Political Science

Review, 114(3), 707–723.

Abramitzky, R., Boustan, L., & Eriksson, K. (2020). Do immigrants assimilate more

slowly today than in the past? American Economic Review: Insights, 2(1), 125–41.

Abramitzky, R., Boustan, L. P., & Eriksson, K. (2014). A nation of immigrants: As-

similation and economic outcomes in the age of mass migration. Journal of Political

Economy, 122(3), 467–506.

Advani, A. & Reich, B. (2015). Melting pot or salad bowl: the formation of heterogeneous

communities. Technical report, IFS Working Papers.

Alesina, A., Miano, A., & Stantcheva, S. (2018). Immigration and redistribution. Tech-

nical report, National Bureau of Economic Research.

Alesina, A., Tabellini, G., & Trebbi, F. (2017). Is Europe an Optimal Political Area?

Working Paper 23325, National Bureau of Economic Research.

Algan, Y., Mayer, T., & Thoenig, M. (2013). The economic incentives of cultural

transmission: Spatial evidence from naming patterns across france.

Bahar, D., Ozguzel, C., Hauptmann, A., & Rapoport, H. (2019). Migration and post-

conflict reconstruction: The effect of returning refugees on export performance in the

former yugoslavia.

Barsbai, T., Rapoport, H., Steinmayr, A., & Trebesch, C. (2017). The effect of labor

migration on the diffusion of democracy: evidence from a former soviet republic.

American Economic Journal: Applied Economics, 9(3), 36–69.

Battisti, M., Giesing, Y., & Laurentsyeva, N. (2019). Can job search assistance improve

the labour market integration of refugees? evidence from a field experiment. Labour

Economics, 61, 101745.

31

Battisti, M., Peri, G., & Romiti, A. (2016). Dynamic effects of Co-Ethnic networks

on immigrants’ economic success. Technical report, National Bureau of Economic

Research.

Becker, S. O. & Ferrara, A. (2019). Consequences of forced migration: A survey of

recent findings. Labour Economics, 59, 1–16.

Bencek, D. & Strasheim, J. (2016). Refugees welcome? a dataset on anti-refugee violence

in germany. Research & Politics, 3(4), 2053168016679590.

Bertrand, M. & Kamenica, E. (2018). Coming apart? Cultural distances in the United

States over time. Technical report, National Bureau of Economic Research.

Bisin, A. & Tura, G. (2019). Marriage, Fertility, and Cultural Integration in Italy.

Technical report, National Bureau of Economic Research.

Bisin, A. & Verdier, T. (2001). The economics of cultural transmission and the dynamics

of preferences. Journal of Economic theory, 97(2), 298–319.

Bleakley, H. & Chin, A. (2010). Age at arrival, english proficiency, and social assimilation

among us immigrants. American Economic Journal: Applied Economics, 2(1), 165–92.

Borjas, G. J. (1985). Assimilation, changes in cohort quality, and the earnings of immi-

grants. Journal of labor Economics, 3(4), 463–489.

Brell, C., Dustmann, C., & Preston, I. (2020). The labor market integration of refugee

migrants in high-income countries. Journal of Economic Perspectives, 34(1), 94–121.

Brucker, H., Rother, N., & Schupp, J. (2016). IAB-BAMF-SOEP-Befragung von

Gefluchteten: Uberblick und erste Ergebnisse. Number 116. DIW Berlin: Politik-

beratung kompakt.

Bundeswahlleiter (2020). Ergebnisse der Bundestagswahl 2017.

Busch, C., Krueger, D., Ludwig, A., Popova, I., & Iftikhar, Z. (2020). Should germany

have built a new wall? macroeconomic lessons from the 2015-18 refugee wave. Journal

of Monetary Economics.

Card, D., Dustmann, C., & Preston, I. (2012). Immigration, wages, and compositional

amenities. Journal of the European Economic Association, 10(1), 78–119.

32

Ceriani, L. & Verme, P. (2018). Risk preferences and the decision to flee conflict. The

World Bank.

Cha, S.-H. (2007). Comprehensive survey on distance/similarity measures between prob-

ability density functions. City, 1(2), 1.

Clarke, D., Romano, J. P., & Wolf, M. (2019). The romano-wolf multiple hypothesis

correction in stata.

Dahlberg, M., Edmark, K., & Lundqvist, H. (2012). Ethnic diversity and preferences

for redistribution. Journal of Political Economy, 120(1), 41–76.

Deole, S. & Huang, Y. (2020). How do new immigration flows affect existing immigrants?

Evidence from the refugee crisis in Germany. GLO Discussion Paper 579.

Desmet, K., Ortuno-Ortın, I., & Wacziarg, R. (2017). Culture, ethnicity, and diversity.

American Economic Review, 107(9), 2479–2513.

Desmet, K. & Wacziarg, R. (2018). The cultural divide. Technical report, National

Bureau of Economic Research.

Edin, P.-A., Fredriksson, P., & Aslund, O. (2003). Ethnic enclaves and the economic

success of immigrants—evidence from a natural experiment. The quarterly journal of

economics, 118(1), 329–357.

El Bialy, N., Nicklisch, A., & Voigt, S. (2017). Risk-taking, trust, and traumatization

of refugees in germany–a field experiment.

Enos, R. D. (2014). Causal effect of intergroup contact on exclusionary attitudes. Pro-

ceedings of the National Academy of Sciences, 111(10), 3699–3704.

Eriksson, K. (2020). Ethnic enclaves and immigrant outcomes: Norwegian immigrants

during the age of mass migration. European Review of Economic History, 24(3), 427–

446.

Fernandez, R. & Fogli, A. (2009). Culture: An empirical investigation of beliefs, work,

and fertility. American Economic Journal: Macroeconomics, 1(1), 146–77.

Fouka, V. (2019). How do immigrants respond to discrimination? the case of germans

in the us during world war i. American Political Science Review, 113(2), 405–422.

33

Fouka, V. (2020). Backlash: The unintended effects of language prohibition in us schools

after world war i. The Review of Economic Studies, 87(1), 204–239.

Fouka, V., Mazumder, S., & Tabellini, M. (2021). From immigrants to americans: Race

and assimilation during the great migration. CEPR Working Paper.

Gagliarducci, S. & Tabellini, M. (2021). Faith and assimilation: Italian immigrants in

the us.

Gehrsitz, M. & Ungerer, M. (2017). Jobs, crime, and votes: A short-run evaluation of

the refugee crisis in germany.

Giavazzi, F., Iglhaut, F., Lemoli, G., & Rubera, G. (2020). Terrorist Attacks, Cultural

Incidents and the Vote for Radical Parties: Analyzing Text from Twitter. Technical

report, National Bureau of Economic Research.

Giuliano, P. & Tabellini, M. (2020). The Seeds of Ideology: Historical Immigration

and Political Preferences in the United States. Technical report, National Bureau of

Economic Research.

Goebel, J., Grabka, M. M., Liebig, S., Kroh, M., Richter, D., Schroder, C., &

Schupp, J. (2019). The German Socio-Economic Panel (SOEP). Jahrbucher fur Na-

tionalokonomie und Statistik, 239(2), 345–360.

Gordon, M. M. (1964). Assimilation in American life: The role of race, religion, and

national origins. Oxford University Press on Demand.

Halla, M., Wagner, A. F., & Zweimuller, J. (2017). Immigration and voting for the far

right. Journal of the European Economic Association, 15(6), 1341–1385.

Hangartner, D., Dinas, E., Marbach, M., Matakos, K., & Xefteris, D. (2019). Does

exposure to the refugee crisis make natives more hostile? American Political Science

Review, 113(2), 442–455.

Hilbig, H. & Riaz, S. (2020). Freedom of movement restrictions inhibit the social inte-

gration of refugees.

Kuhne, S., Jacobsen, J., & Kroh, M. (2019). Sampling in Times of High Immigration:

The Survey Process of the IAB-BAMF-SOEP Survey of Refugees. Survey Methods:

Insights from the Field.

34

Lazear, E. P. (1999). Culture and language. Journal of political Economy, 107(S6),

S95–S126.

Marten, L., Hainmueller, J., & Hangartner, D. (2019). Ethnic networks can foster the

economic integration of refugees. Proceedings of the National Academy of Sciences,

116(33), 16280–16285.

Meng, X. & Gregory, R. G. (2005). Intermarriage and the economic assimilation of

immigrants. Journal of Labor economics, 23(1), 135–174.

Paterson, R. J. & Neufeld, R. W. (1987). Clear danger: Situational determinants of the

appraisal of threat. Psychological bulletin, 101(3), 404.

Phillimore, J. (2020). Refugee-integration-opportunity structures: shifting the focus

from refugees to context. Journal of Refugee Studies.

Rapoport, H., Sardoschau, S., & Silve, A. (2020). Migration and Cultural Change.

Working Papers 2020-10, CEPII.

Reitz, J. G. (2002). Host societies and the reception of immigrants: Research themes,

emerging theories and methodological issues. International migration review, 36(4),

1005–1019.

Romano, J. P. & Wolf, M. (2005a). Exact and Approximate Stepdown Methods for Mul-

tiple Hypothesis Testing. Journal of the American Statistical Association, 100(469),

94–108.

Romano, J. P. & Wolf, M. (2005b). Stepwise Multiple Testing as Formalized Data

Snooping. Econometrica, 73(4), 1237–1282.

Romano, J. P. & Wolf, M. (2016). Efficient computation of adjusted p-values for

resampling-based stepdown multiple testing. Statistics & Probability Letters, 113,

38–40.

Rumbaut, R. G. (2008). Reaping what you sow: Immigration, youth, and reactive

ethnicity. Applied development science, 12(2), 108–111.

Saavedra, M. H. (2018). Kenji or kenneth? pearl harbor and japanese-american assimi-

lation. Pearl Harbor and Japanese-American Assimilation (July 23, 2018).

Statistik der Bundesagentur fur Arbeit, BA (2020). Beschaftigtenstatistik.

35

Steinmayr, A. (2020). Contact versus exposure: Refugee presence and voting for the

far-right. Review of Economics and Statistics, (pp. 1–47).

36

7 Graphs

Figure 1. Monthly Asylum Applications in Germany (total in thousands)

Notes: numbers are taken from Eurostat, which base their information on German statistical offices reports to them. Wecount adult men and women from outside the EU-28, who may have also applied for asylum in other EU countries. These areapplications and not granted asylum status.

Figure 2. Comparison of refugee assignment quotas and actual refugee allocation acrossGerman states

Notes: numbers are taken from the German Statistical Office (Statistisches Bundesamt, Genesis Tab-12531-0025 and Bunde-sanzeiger (2016)) and shows the share of refugees assigned and actually allocated to German state in the year 2016.

37

Figure 3. Average Reciprocity of Locals (NUTS2-Level)

Figure 4. Average level of Threat (NUTS2-Level)

Notes: Data from Bundeswahlleiter (2020), Bencek & Strasheim (2016) and German Socio-Economic Panel. The left mapshows the regional distribution of threat, calculated as the first principal component over right-wing vote shares (AfD-2017,NPD-2013) and hate crimes against refugees in 2014 and 2015 (z-standardized at the NUTS2-level). Fig. A2 shows the regionaldistribution of the 3 components separately. The right map shows the conditional regional distribution in terms of the residualsof a regression of the first threat principal component on 16 Federal State dummies (z-standardized at the NUTS2-level).

38

Figure 6. Assimilation by Mobility (coefficients for dummy of movers among refugees)

Notes: figure include individual-level controls as well as district fixed effects as in our main regression. Bars denote the 95 %confidence interval. Distance relative to first (current) region for stayers (movers) for index and all sub-dimensions. Moverslive in another NUTS-2 region at the time of the interview than they were assigned to. The figure shows the coefficients ofa dummy variable indicating stayers (=0) and movers (=1) in a regression of the Euclidean similarity of refugees to localsresiding in the same NUTS2 region.

Figure 5. Cultural Similarity between Refugees and Locals over time (Index)

Notes: Inverse Euclidean distance over time. Upper and lower line denote the 95 % confidence interval.

39

Figure 7. Simulation of assimilation effort trajectories by threat environment

Notes: Based on our baseline specification in column 7 of Table 4, we predict assimilation effort trajectories of refugees overtime. On the y-axis we plot the level of cultural similarity (Euclidean index) and on the x-axis, we plot months since arrival.We add the effect of the interaction terms between MSA and MSAS with the threat environment, specifically vote for the AfDin 2017, vote for the NPD in 2013, hate-crimes against refugees and openness of locals.

40

41

8 Tables Table 1. Descriptive Statistics - cultural dimensions

All Threat below median Threat above median

mean sd N min max mean sd N min max mean sd N min max

Refugees

Euclidean cultural similarity -1.74 0.32 12,681 -5.76 -0.86 -1.73 0.32 6,066 -3.43 -0.86 -1.75 0.32 6,615 -5.76 -0.89

Risk preferences (0 low - 10 high) 3.94 3.42 12,122 0 10 3.96 3.37 5,793 0 10 3.93 3.46 6,329 0 10

Negative reciprocity (1 low - 7 high) 1.77 1.26 6,399 1 7 1.77 1.26 3,075 1 7 1.76 1.26 3,324 1 7

Positive reciprocity (1 low - 7 high) 6.68 0.62 6,526 1 7 6.70 0.61 3,159 2 7 6.66 0.63 3,367 1 7

Positive self-attitude (1 disagree - 7 agree) 6.29 1.19 6,333 1 7 6.32 1.16 3,075 1 7 6.26 1.22 3,258 1 7

General trust (1 low - 4 high) 2.17 0.59 3,417 1 4 2.21 0.61 1,678 1 4 2.14 0.57 1,739 1 4

Locus of control (1 low - 7 high) 4.46 0.86 3,884 1 7 4.47 0.86 1,666 1 7 4.46 0.87 2,218 1 7

Social inclusion (1 incl. - 5 excl.) 2.57 1.09 6,748 1 5 2.54 1.09 3,255 1 5 2.60 1.10 3,493 1 5

Society exploit-selfish (=1), fair-helpful (=2) 1.57 0.43 3,307 1 2 1.61 0.43 1,626 1 2 1.53 0.44 1,681 1 2

Interest in politics (1 not at all - 4 very strong) 1.66 0.87 12,525 1 4 1.70 0.88 5,993 1 4 1.63 0.86 6,532 1 4

Leisure and cultural activ. (1 never - 5 daily) 1.78 0.63 8,273 1 4 1.80 0.64 4,123 1 4 1.75 0.62 4,150 1 4

Satisfaction with life, health, flat (0 low - 10 high) 7.23 1.94 12,681 0 10 7.24 1.93 6,066 0 10 7.22 1.94 6,615 0 10

Worries: econ., health (1 low - 3 high) 1.83 0.58 12,607 1 3 1.83 0.59 6,037 1 3 1.84 0.58 6,570 1 3

Natives

Euclidean cultural similarity -1.37 0.26 22,335 -3.74 -0.59

Risk preferences (0 low - 10 high) 4.86 2.26 27,903 0 10 4.83 2.29 12,277 0 10 4.88 2.23 15,626 0 10

Negative reciprocity (1 low - 7 high) 3.04 1.42 18,720 1 7 3.01 1.43 7,779 1 7 3.07 1.42 10,941 1 7

Positive reciprocity (1 low - 7 high) 5.84 0.91 18,764 1 7 5.83 0.91 7,803 1 7 5.84 0.91 10,961 1 7

Positive self-attitude (1 disagree - 7 agree) 5.58 1.28 18,817 1 7 5.58 1.29 7,825 1 7 5.59 1.26 10,992 1 7

General trust (1 low - 4 high) 2.37 0.53 25,756 1 4 2.41 0.53 11,315 1 4 2.34 0.54 14,441 1 4

Locus of control (1 low - 7 high) 4.61 0.73 18,868 1 7 4.63 0.74 7,845 1 7 4.60 0.72 11,023 1 7

Social inclusion (1 incl. - 5 excl.) 2.02 0.75 25,767 1 5 2.00 0.75 11,319 1 5 2.03 0.76 14,448 1 5

Society exploit-selfish (=1), fair-helpful (=2) 1.52 0.42 25,618 1 2 1.54 0.42 11,255 1 2 1.50 0.43 14,363 1 2

Interest in politics (1 not at all - 4 very strong) 2.35 0.81 20,773 1 4 2.40 0.81 8,894 1 4 2.31 0.81 11,879 1 4

Leisure and cultural activ. (1 never - 5 daily) 2.14 0.63 25,790 1 5 2.19 0.62 11,326 1 5 2.10 0.63 14,464 1 4

Satisfaction with life, health, flat (0 low - 10 high) 7.24 1.48 26,772 0 10 7.34 1.47 11,732 0 10 7.17 1.48 15,040 0 10

Worries: econ., health (1 low - 3 high) 1.91 0.61 27,886 1 3 1.88 0.62 12,280 1 3 1.94 0.61 15,606 1 3

Table 1 shows descriptive statistics including mean, standard deviation, number of observation, minimum and maximum values for all cultural variables, including the Euclideancultural similarity index and all it’s components for both locals and refugees as well as low and high threat NUTS2-regions. Threat regions are classified at NUTS2-level as belowor above the median in terms of the first threat principal component (described in section 3.1). ** denotes statistically significant differenes at the 95 % level. Data come fromGerman Socio-Economic Panel (SOEP) and the IAB-BAMF-SOEP Survey of Refugees.

42

Table 2. Descriptive Statistics - individual controls

mean sd N min max mean sd N min max

Refugees Locals

Individual-level controls

Months since arrival to Germany 29.56 16.23 12,680 0 392

Age (in years) 34.09 10.22 12,680 18 66 51.08 18.55 247,013 17 105

Gender: female 0.38 0.49 12,680 0 1 0.51 0.50 260,278 0 1

Years of work exp. before arrival 7.35 9.23 11,919 0 48

Finished educ. (refugees: before arrival): No 0.45 0.50 12,621 0 1 0.11 0.31 272,665 0 1

with school leaving certificate 0.24 0.43 12,621 0 1 0.01 0.10 272,665 0 1

with secondary school leaving certificate 0.32 0.47 12,621 0 1 0.88 0.33 272,665 0 1

No Partner 0.33 0.47 12,680 0 1

lives in household 0.58 0.49 12,680 0 1

elsewhere in Germany 0.01 0.11 12,680 0 1

not in Germany 0.06 0.24 12,680 0 1

Nationality: Germany 0.00 0.00 12,680 0 0 0.91 0.29 271,010 0 1

Syria 0.52 0.50 12,680 0 1 0.00 0.02 271,010 0 1

Afghanistan 0.13 0.33 12,680 0 1 0.00 0.02 271,010 0 1

Iraq 0.13 0.34 12,680 0 1 0.00 0.03 271,010 0 1

Africa 0.08 0.28 12,680 0 1 0.00 0.05 271,010 0 1

West Balkan 0.03 0.16 12,680 0 1 0.01 0.10 271,010 0 1

Poland 0.00 0.00 12,680 0 0 0.00 0.07 271,010 0 1

Turkey 0.00 0.00 12,680 0 0 0.02 0.14 271,010 0 1

Italy 0.00 0.00 12,680 0 0 0.01 0.10 271,010 0 1

Other 0.11 0.32 12,680 0 1 0.05 0.21 271,010 0 1

Table 2 shows descriptive statistics including mean, standard deviation, number of observation, minimum and maximumvalues for individual survey variables. Data come from Socio-Economic Panel (SOEP) and the IAB-BAMF-SOEP Survey ofRefugees. Descriptive stats for locals (right panel) are weighted.

43

Table 3. Descriptive Statistics - Other Data Sources

mean sd N min max

District-level controls, measured in 2012

Unemployment rate 6.84 2.97 12,680 1.2 15.9

Population density 955.06 1104.80 12,680 38 4,468

Share of refugees 0.75 0.37 12,680 0.06 2.4

Mediator variables (at NUTS2-level):

Local-national cultural distance (Euclidean) 0.46 0.21 12,680 0.20 1.35

Local cultural dispersion (within-region sd) 0.78 0.02 12,680 0.74 0.83

Right-wing vote (AfD 2017) (percentage) 12.31 4.52 12,680 7.80 29.66

Right-wing vote (NPD 2013) (percentage) 1.26 0.70 12,680 0.50 3.65

Hate crimes against refugees (per 100k inhabitants) 1.95 2.23 12,680 0.16 12.70

Share of refugees 2.09 0.50 12,680 1.14 4.24

Immigrants from origin region 0.74 0.75 11,170 0.05 10.66

Refugees’ cultural similarity -1.75 0.13 12,680 -1.97 -1.45

Employment rate of immigrants from origin region 17.11 6.60 11,170 1.30 54.19

Gross wage of immigrants from origin region (Euro) 1,122 217 11,988 426 1,826

Table 3 shows descriptive statistics including mean, standard deviation, number of observation,

minimum and maximum values for all non-survey data. Unemployment rate, population density

and the share of refugees is measured in December-2012. N denotes person-year observations,

differences are due to missing values.

44

Table 4. Assimilation effort: months since arrival and cultural similarity to locals

Dep. Var.: Euclidean Cultural Similarity Index(1) (2) (3) (4) (5) (6) (7)

Months since arrival 0.140∗∗∗ 0.136∗∗∗ 0.126∗∗∗ 0.127∗∗∗ 0.133∗∗∗ 0.137∗∗∗ 0.151∗∗∗

(MSA) (0.035) (0.035) (0.035) (0.035) (0.033) (0.033) (0.034)

Months since arrival, squared -0.001∗∗∗ -0.001∗∗∗ -0.000∗∗∗ -0.000∗∗∗ -0.001∗∗∗ -0.001∗∗∗ -0.001∗∗∗

(MSAS) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

ControlsIndividual-level No No Yes Yes Yes Yes YesDistrict-level No No No Yes Yes Yes Yes

Fixed EffectsComposition Yes Yes Yes Yes Yes Yes YesSurvey year No Yes Yes Yes Yes Yes YesFederal-State No No No No Yes No NoNUTS-2 No No No No No Yes NoDistrict No No No No No No Yes

Person-year observations 12,681 12,681 12,681 12,681 12,681 12,681 12,681Person observations 6,867 6,867 6,867 6,867 6,867 6,867 6,867R2 adjusted 0.181 0.181 0.196 0.200 0.225 0.231 0.252

Table 4 shows our main results, successively introducing control variables at the individual and district level as wellas fixed effects for the survey year and geographic fixed effects. We also include composition fixed effects that accountpotential changes in the survey questionnaire over time. We also report the squared term for months since arrival(MSAS). Cultural similarity measured as the inverse Euclidean distance to non-refugees in the NUTS2 region thatthe refugee was assigned to. The index includes: risk, reciprocity, leisure and cultural activities, satisfaction, wor-ries, political interest, locus of control, social inclusion, self-attitude, trust, egoistic-fair society. Individual controlsinclude: gender, age, age squared, partnership (no partner, partner lives in household, elsewhere in Germany or notin Germany), years of work experience at entry and level of education at entry. District-level controls include: unem-ployment rate, population density and share of asylum seekers in Dec-2012 (interacted with survey year dummies).Omitted covariates are shown in Table A2 and a Canberra-modified version of the Euclidean Similarity Index inTable B1. Coefficients and SE multiplied by 100 for presentation. Positive coefficients indicate a reduction in culturaldistance. Standard errors in parentheses clustered at person-level. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

45

Table 5. Assimilation success: months since arrival, employment and log wage

(1) (2) (3) (4) (5) (6) (7)Outcome: Employed at time of the interview

Months since arrival 0.888∗∗∗ 0.880∗∗∗ 0.818∗∗∗ 0.814∗∗∗ 0.820∗∗∗ 0.818∗∗∗ 0.808∗∗∗

(MSA) (0.044) (0.060) (0.057) (0.056) (0.056) (0.057) (0.057)

Months since arrival, squared -0.002∗∗∗ -0.002∗∗∗ -0.002∗∗∗ -0.002∗∗∗ -0.002∗∗∗ -0.002∗∗∗ -0.002∗∗∗

(MSAS) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Person-year observations 12,681 12,681 12,681 12,681 12,681 12,681 12,681Person observations 6,867 6,867 6,867 6,867 6,867 6,867 6,867R2 adjusted 0.073 0.073 0.155 0.160 0.161 0.163 0.177

Outcome: Log wage at time of the interview (among employed)

Months since arrival 1.362∗∗∗ 0.889∗∗∗ 0.683∗∗ 0.669∗∗ 0.674∗∗ 0.726∗∗ 0.655∗∗

(MSA) (0.236) (0.278) (0.285) (0.288) (0.288) (0.286) (0.312)

Months since arrival, squared -0.004∗∗∗ -0.002∗∗ -0.001 -0.001 -0.001 -0.001 -0.001(MSAS) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)

Person-year observations 2,171 2,171 2,171 2,171 2,171 2,171 2,171Person observations 1,618 1,618 1,618 1,618 1,618 1,618 1,618R2 adjusted 0.022 0.030 0.095 0.098 0.098 0.105 0.143

ControlsIndividual-level No No Yes Yes Yes Yes YesDistrict-level No No No Yes Yes Yes Yes

Fixed EffectsSurvey year No Yes Yes Yes Yes Yes YesFederal-State No No No No Yes No NoNUTS-2 No No No No No Yes NoDistrict No No No No No No Yes

Table 5 shows our main results, successively introducing control variables at the individual and district level as wellas fixed effects for the survey year and geographic fixed effects. We also report the squared term for months sincearrival (MSAS). Assimilation success is measured as the employment status and the log wage among employed at thetime of the interview. Individual controls include: gender, age, age squared, partnership (no partner, partner livesin household, elsewhere in Germany or not in Germany), years of work experience at entry and level of education atentry. District-level controls in Dec-2012 include: unemployment rate, population density and share of asylum seekers(interacted with survey-year dummies). Coefficients and SE multiplied by 100 for presentation. Positive coefficientsindicate a reduction in cultural distance. Standard errors in parentheses clustered at person-level. + p < 0.15, ∗

p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

46

Table 6. Assimilation effort under threat

Dep. Var.: Euclidean Cultural Similarity Index

(1) (2) (3) (4)first threat right-wing vote right-wing vote hate crimes

PC AfD 2017 NPD 2013 against refugees

Panel A: Linear specification

Months since arrival 0.053∗∗ 0.052∗∗ 0.053∗∗ 0.053∗∗

(0.021) (0.021) (0.021) (0.022)

Months since arrival × mediator 0.014 0.017 0.020 0.027(0.014) (0.020) (0.022) (0.028)

Person-Year observations 12,680 12,680 12,680 12,680Person observations 6,866 6,866 6,866 6,866R2 adjusted 0.268 0.267 0.267 0.268

Panel B: Polynomial specification (2nd order)

Months since arrival 0.173∗∗∗ 0.170∗∗∗ 0.170∗∗∗ 0.171∗∗∗

(0.037) (0.036) (0.036) (0.036)

Months since arrival × mediator 0.073∗∗∗ 0.097∗∗∗ 0.109∗∗∗ 0.134∗∗∗

(0.020) (0.030) (0.032) (0.036)

Person-Year observations 12,680 12,680 12,680 12,680Person observations 6,866 6,866 6,866 6,866R2 adjusted 0.269 0.269 0.269 0.269

ControlsIndividual-level Yes Yes Yes YesDistrict-level Yes Yes Yes Yes

Fixed EffectsComposition Yes Yes Yes YesSurvey year Yes Yes Yes YesDistrict Yes Yes Yes Yes

Table 6 shows baseline specifications with interaction between our main exogenous variable, monthssince arrival (MSA), and other mediating variables at the local level focusing on features of the non-refugee community in the same NUTS2 region. All specifications in Panel A include MSA but notMSAS. Panel B also includes the interaction term between months since arrival squared (MSAS) andthe mediating variable (not reported). In column 1, we use the first PC threat index. In column 2and 3, we use the right-wing vote share for the AfD in 2017 and the NPD in 2013. In column 4, wetake geo-located hate crimes against refugees between 2014 and 2015 from Bencek & Strasheim (2016).Standard errors in parentheses, clustered at person-level. Coefficients and SE multiplied by 100 forpresentation. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

47

Table 7. Assimilation success under threat

Dep. Var.: employed/log wage at time of interview

first threat PC AfD 2017 NPD 2013 hate crimes

(1) (2) (3) (4) (5) (6) (7) (8)empl. wage empl. wage empl. wage empl. wage

Panel A: Linear specification

MSA 0.495∗∗∗ 0.916∗∗∗ 0.497∗∗∗ 0.902∗∗∗ 0.495∗∗∗ 0.915∗∗∗ 0.494∗∗∗ 0.945∗∗∗

(0.044) (0.178) (0.044) (0.177) (0.043) (0.176) (0.044) (0.191)

MSA × mediator -0.054∗ 0.049 -0.053∗ -0.028 -0.056∗ 0.053 -0.049+ 0.182(0.030) (0.230) (0.030) (0.182) (0.032) (0.223) (0.030) (0.320)

Person-year observations 12,681 2,279 12,681 2,279 12,681 2,279 12,681 2,279Person observations 6,867 1,679 6,867 1,679 6,867 1,679 6,867 1,679R2 adjusted 0.172 0.137 0.172 0.137 0.172 0.137 0.172 0.137

Panel B: Polynomial specification (2nd order)

MSA 0.829∗∗∗ 2.206∗∗∗ 0.833∗∗∗ 2.145∗∗∗ 0.827∗∗∗ 2.233∗∗∗ 0.826∗∗∗ 2.211∗∗∗

(0.052) (0.513) (0.052) (0.507) (0.052) (0.516) (0.054) (0.505)

MSA × mediator -0.015 0.316 -0.009 0.053 -0.004 0.452 -0.023 0.426(0.038) (0.587) (0.037) (0.548) (0.039) (0.564) (0.038) (0.594)

Person-year observations 12,681 2,279 12,681 2,279 12,681 2,279 12,681 2,279Person observations 6,867 1,679 6,867 1,679 6,867 1,679 6,867 1,679R2 adjusted 0.178 0.140 0.178 0.140 0.178 0.140 0.178 0.140

ControlsIndividual-level Yes Yes Yes Yes Yes Yes Yes YesDistrict-level Yes Yes Yes Yes Yes Yes Yes Yes

Fixed EffectsSurvey year Yes Yes Yes Yes Yes Yes Yes YesDistrict Yes Yes Yes Yes Yes Yes Yes Yes

Table 7 shows baseline specifications with interaction between our main exogenous variable, months since arrival (MSA),and other mediating variables at the local level focusing on political and socio-psychological features of the non-refugeecommunity in the same NUTS2 region. All specifications in Panel A include MSA but not MSAS. Panel B also includesthe interaction term between months since arrival squared (MSAS) and the mediating variable (not reported). Incolumns 1 and 2, we use the first PC threat index. In columns 3 and 4, we use the NUTS2-level vote share for theanti-immigration party AfD in the year 2017. In columns 5 and 6, we take the extreme right party NPD in the year2013. In columns 7 and 8, we use geo-located hate crimes against refugees between 2014 and 2015 from Bencek &Strasheim (2016). Standard errors in parentheses, clustered at person-level. Coefficients and SE multiplied by 100 forpresentation. + p < 0.15, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

48

Table 8. Assimilation effort and success- refugee and immigrant networks

(1) (2) (3) (4) (5) (6)Mediator: Share of refugees Employment of immigrantsOutcome: Effort Success Success Effort Success Success

(CS) (empl.) (log wage) (CS) (empl.) (log wage)

Panel A: Linear specification

Months since arrival 0.048∗∗ 0.493∗∗∗ 0.412∗∗∗ 0.050∗∗ 0.517∗∗∗ 0.416∗∗∗

(0.023) (0.043) (0.119) (0.024) (0.047) (0.127)

Months since arrival × mediator -0.052∗∗∗ 0.002 0.080 -0.004 0.015 -0.031(0.019) (0.032) (0.138) (0.018) (0.034) (0.121)

Person-year observations 12,681 12,681 2,171 12,117 12,117 2,046Person observations 6,867 6,867 1,618 6,534 6,534 1,525R2 adjusted 0.252 0.171 0.142 0.251 0.173 0.145

Panel B: Polynomial specification (2nd order)

Months since arrival 0.163∗∗∗ 0.831∗∗∗ 0.632∗ 0.156∗∗∗ 0.833∗∗∗ 0.597∗

(0.035) (0.050) (0.332) (0.038) (0.059) (0.357)

Months since arrival × mediator -0.106∗∗∗ -0.135∗∗∗ 0.211 -0.044+ 0.017 0.128(0.027) (0.038) (0.280) (0.029) (0.044) (0.287)

Person-year observations 12,681 12,681 2,171 12,117 12,117 2,046Person observations 6,867 6,867 1,618 6,534 6,534 1,525R2 adjusted 0.253 0.178 0.143 0.252 0.178 0.145

ControlsIndividual-level Yes Yes Yes Yes Yes YesDistrict-level Yes Yes Yes Yes Yes Yes

Fixed EffectsComposition Yes No No Yes No NoSurvey year Yes Yes Yes Yes Yes YesDistrict Yes Yes Yes Yes Yes Yes

Table 8 shows baseline specifications with interaction between our main exogenous variable, months sincearrival (MSA), and the mediating variables share of refugees residing in the NUTS2 region of assignment fromthe same origin region (columns 1 to 3) and employment of immigrants (columns 4 to 6) at the local level(z-standardized). We define origin region along 5 categories: MENA, Sub-Saharan Africa, West Balkans,former USSR, and Afghanistan. All specifications in Panel A include MSA but not MSAS. Panel B alsoincludes the interaction term between months since arrival squared (MSAS) and the mediating variable (notreported). In columns 1 and 4, we calculate from our survey sample the average cultural similarity of allrefugees assigned to the same NUTS2 region. In column 2 and 5, we take the employment rate and in columns3 and 6 the log wage of all immigrants from the same origin region residing in the assigned NUTS2 region.Standard errors in parentheses, clustered at person-level. Coefficients and standard errors multiplied by 100for presentation. + p < 0.15, ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

49

Appendix A. Additional Figures and Tables

Figure A1. Visual Presentation of the Conceptual Framework

Notes: Visual presentation of the theoretical framework. The blue line represent the optimal assimilation path for refugeesin a neutral environment, with a threshold level of acquired assimilation C and timing of successful assimilation at t∗.The red line represent the adjusted optimal assimilation path of refugees in a threat environment with a higher thresholdlevel of required assimilation C′. In the absence of adjusted assimilation in threat environments, refugees enter the labor

market at t′∗, with adjusted assimilation refugees enter at t

′′∗

50

Figure A2. Average level of Threat, single components (NUTS2-Level)

Notes: Data from Bundeswahlleiter (2020), Bencek & Strasheim (2016) and German Socio-Economic Panel. The Figure shows the regional distribution of natives’threat-dimension across NUTS2-regions. The 3 plotted components constitute the first threat principal component that is plotted in Fig. 4.

51

Figure A3. Cultural Similarity between Refugees and Locals over NUTS2 regions (Index)

Notes: Inverse Euclidean distance of refugees to locals, averaged at NUTS2-level.

52

Figure A4. Refugee outcomes over time since arrival (conditional)

Notes: spikes denote the 95 % confidence intervals. The figure shows the coefficients of a factor recoded version of monthssince arrival) in a regression of refugee outcomes on the same set of remaining covariates as in our main regression (col. 6, 7in Table 4). Source: IAB-BAMF-SOEP Survey of Refugees 2016-2018.53

Figure A5. Euclidean distance between natives’ regional and national culture

Notes: This map shows the difference between natives’ local and national culture at the NUTS2-level. We calculate thedifference as the Euclidean distance between mean deviations of residents of the NUTS2 region from whole Germany.The selection of questions is equivalent to the Euclidean index between refugees and locals. See Table A1 for the list ofquestions.

54

Table A1. Survey Questions used for Aggregate Euclidean Index

Outcome Variables Survey YearCategory Question Answer Refugees LocalsRisk In general, are you someone who is ready to take risks or do you try to avoid risks? 0 risk averse - 10 fully prepared to take risks 2017 2012Positive Reci-procity

If someone does me a favour, I am willing to reciprocate it 1 Absolutely does not apply - 7 Fully applies 2017 2010

I make particular effort to help someone who has previously helped me. 1 Absolutely does not apply - 7 Fully applies 2017 2010I am prepared to incur costs myself to help someone who has previously helped me. 1 Absolutly does not apply - 7 Fully applies 2017 2010

Negative Reci-procity

If someone does me a serious wrong, I will get my own back at any price at the next oppor-tunity.

1 Absolutely does not apply - 7 Fully applies 2017 2010

If somebody puts me in a difficult position, I will do the same to them. 1 Absolutely does not apply - 7 Fully applies 2017 2010If someone insults me, I will insult them. 1 Absolutely does not apply - 7 Fully applies 2017 2010

Leisure Activi-ties

How often do you go to eat or drink in a cafe, restaurant or bar? 1 Never - 5 Daily 2017 2013

Artistic and musical activities (painting, music, photography, theater, dance) 1 Never - 5 Daily 2017 2013Taking part in sports 1 Never - 5 Daily 2017 2013Going to sporting events 1 Never - 5 Daily 2017 2013Going to the cinema, pop concerts, dance events, clubs 1 Never - 5 Daily 2017 2013Going to cultural events such as opera, classical concerts, theater, exhibitions 1 Never - 5 Daily 2017 2013

Satisfaction How satisfied are you currently with your life in general? 0 Completely dissatisfied - 10 Completely satisfied 2017 2012How satisfied are you with your current health? 0 Completely dissatisfied - 10 Completely satisfied 2017 2012How satisfied are you in general with your current living arrangements? 0 Completely dissatisfied - 10 Completely satisfied 2017 2012

Worries Are you worried about your own economic situation? 1 No, no worry - 3 Yes, big worry 2016, 2017, 2018 2012Are you worried about your health? 1 No, no worry - 3 Yes, big worry 2016, 2017, 2018 2012Are you worried about xenophobia and racial hatred in Germany? 1 No, no worry - 3 Yes, big worry 2016, 2017, 2018 2012

Politics Once spoken in general terms: How interested are you in politics 1 Not at all - 4 Very strong 2016, 2017, 2018 2012Locus of Control How my life goes depends on me 1 Absolutely does not apply - 7 Fully applies 2016 2010

Compared to other people, I have not achieved what I deserve 1 Absolutely does not apply - 7 Fully applies 2016 2010What a person achieves in life is above all a question of fate or luck 1 Absolutely does not apply - 7 Fully applies 2016 2010If a person is socially or politically active, he/she can have an effect on social conditions 1 Absolutely does not apply - 7 Fully applies 2016 2010I frequently have the experience that other people have a controlling influence over my life 1 Absolutely does not apply - 7 Fully applies 2016 2010One has to work hard in order to succeed 1 Absolutely does not apply - 7 Fully applies 2016 2010If I run up against difficulties in life, I often doubt my own abilities 1 Absolutely does not apply - 7 Fully applies 2016 2010The opportunities that I have in life are determined by the social conditions 1 Absolutely does not apply - 7 Fully applies 2016 2010Inborn abilities are more important than any efforts one can make 1 Absolutely does not apply - 7 Fully applies 2016 2010I have little control over the things that happen in my life 1 Absolutely does not apply - 7 Fully applies 2016 2010

Social Inclusion How often do you miss the company of other people? 1 Never - 5 Very often 2016-2018 (Bio) 2013How often do you feel left out? 1 Never - 5 Very often 2016-2018 (Bio) 2013How often do you feel socially isolated? 1 Never - 5 Very often 2016-2018 (Bio) 2013

Self Attitude I have a positive attitude towards myself 1 Absolutely does not apply - 7 Fully applies 2016-2018 (Bio) 2010Trust People can generally be trusted 1 Not at all - 4 Fully agree 2018 2013

Nowadays you can’t rely on anyone 1 Not at all - 4 Fully agree 2018 2013If you are dealing with strangers, it is better to be careful before trusting them 1 Not at all - 4 Fully agree 2018 2013

Egoistic society Do you believe that most people would use you if they had the chance or that they wouldtry to be fair to you?

1 exploit - 2 fair 2018 2013

Would you say that people usually try to be helpful or that they only pursue their owninterest?

1 own interest - 2 helpful 2018 2013

55

Table A2. Assimilation effort: presentation of covariates omitted from Table 4

(1) (2) (3) (4) (5) (6) (7)MSA 0.140∗∗∗ 0.136∗∗∗ 0.126∗∗∗ 0.127∗∗∗ 0.133∗∗∗ 0.137∗∗∗ 0.151∗∗∗

(0.035) (0.035) (0.035) (0.035) (0.033) (0.033) (0.034)MSAS -0.001∗∗∗ -0.001∗∗∗ -0.000∗∗∗ -0.000∗∗∗ -0.001∗∗∗ -0.001∗∗∗ -0.001∗∗∗

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Female (ref. male) -3.258∗∗∗ -3.269∗∗∗ -3.376∗∗∗ -3.306∗∗∗ -3.322∗∗∗

(0.680) (0.676) (0.659) (0.652) (0.643)Age -0.230 -0.251 -0.136 -0.171 -0.117

(0.198) (0.198) (0.192) (0.191) (0.189)Age, squared 0.001 0.001 -0.001 -0.000 -0.001

(0.003) (0.003) (0.003) (0.002) (0.002)Partner: lives in HH (ref. no partner) 0.660 0.648 0.993 1.021 0.936

(0.781) (0.779) (0.756) (0.755) (0.759)lives elsewhere in Germany 1.894 2.087 1.918 2.025 2.832

(2.280) (2.278) (2.251) (2.250) (2.291)not in Germany -4.009∗∗∗ -4.035∗∗∗ -3.715∗∗∗ -3.778∗∗∗ -3.627∗∗∗

(1.350) (1.344) (1.303) (1.298) (1.316)Years of work experience before immigration 0.014 0.012 0.003 0.004 -0.021

(0.050) (0.049) (0.048) (0.048) (0.047)Some school leaving certificate (ref. none) 3.880∗∗∗ 3.900∗∗∗ 3.500∗∗∗ 3.561∗∗∗ 3.271∗∗∗

(0.733) (0.729) (0.708) (0.707) (0.706)Secondary certificate 6.577∗∗∗ 6.650∗∗∗ 7.051∗∗∗ 7.244∗∗∗ 7.375∗∗∗

(0.680) (0.677) (0.656) (0.653) (0.650)Afghanistan (ref. Syria) -0.440 -0.361 -0.239 0.239 -0.337

(0.920) (0.914) (0.896) (0.899) (0.894)Iraq -0.991 -0.825 -0.773 -0.898 -1.184

(0.884) (0.876) (0.869) (0.875) (0.882)Africa 1.358 1.647 2.490∗∗ 2.840∗∗∗ 2.183∗∗

(1.135) (1.135) (1.085) (1.088) (1.106)Western Balkans -3.277∗ -3.504∗ -3.137 -3.165∗ -3.187

(1.919) (1.925) (1.926) (1.910) (1.957)Other/Stateless 0.567 0.684 1.305 1.201 1.429

(0.950) (0.950) (0.935) (0.931) (0.940)Surveyyear: 2017 (ref. 2016) -4.123 -5.354 -4.219 -3.527 -4.134 -3.488

(12.930) (12.359) (12.820) (12.247) (12.305) (12.730)2018 1.660 1.718 1.819 0.731 0.769 0.504

(4.393) (4.357) (4.760) (4.772) (4.795) (4.870)Unemployment 0.412∗∗∗ 0.288 0.159

(0.153) (0.212) (0.235)Surveyyear: 2017 × unemployment -0.022 0.021 0.004 0.114

(0.216) (0.216) (0.217) (0.225)2018 × unemployment 0.523∗∗ 0.521∗∗ 0.510∗∗ 0.557∗∗

(0.213) (0.213) (0.214) (0.222)Pop-density -0.002∗∗∗ -0.001∗ -0.000

(0.001) (0.001) (0.001)Surveyyear 2017 × Pop-density 0.001 0.000 0.000 0.000

(0.001) (0.001) (0.001) (0.001)2018 × Pop-density -0.001 -0.001 -0.001 -0.001

(0.001) (0.001) (0.001) (0.001)Refugees’ share 6.196∗∗∗ 4.115∗∗ 1.852

(1.558) (1.786) (1.881)Surveyyear 2017 × Refugees’ share -2.492 -0.366 -0.157 0.042

(2.194) (2.193) (2.203) (2.273)2018 × Refugees’ share -4.446∗∗ -1.922 -1.562 -1.269

(2.155) (2.153) (2.155) (2.218)Person-year observations 12,681 12,681 12,681 12,681 12,681 12,681 12,681Person observations 6,867 6,867 6,867 6,867 6,867 6,867 6,867R2 adjusted 0.181 0.181 0.196 0.200 0.225 0.231 0.252

Composition Yes Yes Yes Yes Yes Yes YesFederal-State No No No No Yes No NoNUTS-2 No No No No No Yes NoDistrict No No No No No No Yes

Table A2 runs the same analyses as in Table 4 showing all covariates. Positive coefficients indicate a reduction in cultural distance.Cultural similarity index includes: risk, reciprocity, leisure and cultural activities, satisfaction, worries, political interest, locus ofcontrol social inclusion, self-attitude, trust, egoistic-fair society. Standard errors in parentheses clustered at person-level. Regionalcontrols are measured at district-level in Dec-2012. All specifications control for a regressions constant and missing categories incovariates. Coefficients and SE multiplied by 100 for presentation. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

56

Table A3. Cultural convergence by question

(1) (2) (3) (4) (5) (6)Risk Rec-Neg Rec-Pos Self-Att. Trust Locus

MSA 0.083 0.068 0.108∗∗∗ 0.092 -0.029 -0.080(0.149) (0.069) (0.036) (0.080) (0.094) (0.074)

Person-year observations 12,122 6,399 6,526 6,333 3,417 3,884Person observations 6,695 6,399 6,526 6,333 3,417 3,884R2 adjusted 0.079 0.115 0.249 0.076 0.099 0.054

(7) (8) (9) (10) (11) (12)Soc-Incl Ego-Fair-Soc Polit Active Satisf Worries

MSA 0.308∗∗∗ -0.024 0.149∗∗∗ 0.232∗∗∗ 0.301∗∗∗ -0.021(0.083) (0.030) (0.047) (0.040) (0.103) (0.029)

Person-year observations 6,748 3,307 12,525 8,273 12,681 12,607Person observations 6,748 3,307 6,812 5,272 6,867 6,834R2 adjusted 0.059 0.060 0.088 0.113 0.042 0.036

ControlsIndividual-level Yes Yes Yes Yes Yes YesDistrict-level Yes Yes Yes Yes Yes YesMSAS Yes Yes Yes Yes Yes Yes

Fixed EffectsComposition No No No No No NoSurvey year Yes Yes Yes Yes Yes YesDistrict Yes Yes Yes Yes Yes Yes

Table A3 shows our baseline specification by question. Cultural similarity measured as the inverseEuclidean distance to non-refugees in the NUTS2 region that the refugee was assigned to. The index,control variables and fixed effects are the same as in our preferred specification in Table 4 column 7(except dummies for outcome-index composition, as we are looking at specific questions). Positivecoefficients indicate a reduction in cultural distance. Standard errors in parentheses, clustered atperson-level. Coefficients and standard errors multiplied by 100 for presentation. ∗ p < 0.10, ∗∗

p < 0.05, ∗∗∗ p < 0.01. Statistical significance levels adjusted for multiple hypotheses testing bycontrolling the familywise error rate (FWER) using the Romano-Wolf procedure (Clarke et al.,2019; Romano & Wolf, 2016, 2005a,b).

57

Table A4. Mutual interaction of assimilation effort and success under threat

Effort (CS) Success (empl.)(1) (2) (3) (4)

Baseline first threat PC Baseline first threat PC

Panel A: Linear specification

MSA 0.057 0.067+ 0.505∗∗∗ 0.544∗∗∗

(0.048) (0.046) (0.066) (0.071)lag employed 5.080+ 5.312∗

(3.213) (2.806)MSA × mediator × lag employed -0.145∗∗

(0.065)

lagged CS 0.068 -0.741(2.306) (2.385)

MSA × mediator × lagged CS -0.083∗∗

(0.035)

Person-year observations 5,393 5,393 5,393 5,393Person observations 3,785 3,785 3,785 3,785R2 adjusted 0.310 0.311 0.204 0.206

Panel B: Polynomial specification (2nd order)

MSA 0.215∗∗ 0.157+ 0.990∗∗∗ 1.006∗∗∗

(0.092) (0.098) (0.114) (0.122)lag employed 5.750 7.260

(5.427) (5.538)MSA × mediator × lag employed -0.339∗

(0.180)

lagged CS -0.765 -1.384(2.866) (3.096)

MSA × mediator × lagged CS -0.235∗∗

(0.093)Person-year observations 5,393 5,393 5,393 5,393Person observations 3,785 3,785 3,785 3,785R2 adjusted 0.310 0.312 0.209 0.210

ControlsIndividual-level Yes Yes Yes YesDistrict-level Yes Yes Yes Yes

Fixed EffectsComposition Yes Yes No NoSurvey year Yes Yes Yes YesDistrict Yes Yes Yes Yes

Table A4 shows interactions between our main exogenous variable, months since arrival(MSA), and other mediating variables(z-standardized). All specifications in Panel A includeMSA but not MSAS. Panel B also includes the interaction term between months since arrivalsquared (MSAS) and the mediating variable (not reported). We triple interact MSA withlagged cultural similarity and lagged employment. We also include the pairwise interactions(not reported). We use the PCA index for threat in columns 2 and 4. Standard errors inparentheses, clustered at person-level. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Coefficients andstandard errors multiplied by 100 for presentation.

58

Table A5. Assimilation effort under threat - triple interaction with refugee shelter

(1) (2) (3) (4)first threat right-wing vote right-wing vote hate crimes

PC (AFD 2017) (NPD 2013) against refugeesPanel A: Linear specification

Months since arrival 0.034+ 0.034+ 0.035+ 0.033+

(0.022) (0.022) (0.022) (0.022)

Months since arrival × mediator 0.004 0.003 0.001 0.007(0.021) (0.019) (0.020) (0.024)

Months since arrival × mediator 0.138∗∗∗ 0.107∗∗ 0.141∗∗∗ 0.122∗∗

× shared accommodation (0.053) (0.052) (0.052) (0.052)

Person-year observations 12,630 12,630 12,630 12,630Person observations 6,859 6,859 6,859 6,859R2 adjusted 0.254 0.253 0.254 0.254

Panel B: Polynomial specification (2nd order)

Months since arrival 0.141∗∗∗ 0.139∗∗∗ 0.137∗∗∗ 0.140∗∗∗

(0.039) (0.039) (0.039) (0.039)

Months since arrival × mediator 0.070∗∗ 0.058∗ 0.056∗ 0.088∗∗

(0.034) (0.032) (0.033) (0.038)

Months since arrival × mediator 0.260+ 0.167 0.245 0.264× shared accommodation (0.180) (0.143) (0.171) (0.185)

Person-year observations 12,630 12,630 12,630 12,630Person observations 6,859 6,859 6,859 6,859R2 adjusted 0.254 0.254 0.254 0.255

ControlsIndividual-level Yes Yes Yes YesDistrict-level Yes Yes Yes Yes

Fixed EffectsComposition Yes Yes Yes YesSurvey year Yes Yes Yes YesDistrict Yes Yes Yes Yes

Table A5 shows baseline specifications with interaction between our main exogenous variable, monthssince arrival (MSA), and other mediating variables at the local level (z-standardized) focusing on po-litical and socio-psychological features of the non-refugee community in the same NUTS2 region. Allspecifications in Panel A include MSA but not MSAS. Panel B also includes the interaction termbetween months since arrival squared (MSAS) and the mediating variable (not reported). We tripleinteract MSA with the threat environment and shared accommodation. We also include the pairwiseinteractions (not reported). We use the PCA index for threat in column 1. In column 2, we use theNUTS2-level vote share for the anti-immigration party AfD in the year 2017. In column 3, we take theextreme right party NPD in the year 2013. In column 4, we use geo-located hate crimes against refugeesbetween 2014 and 2015 from Bencek & Strasheim (2016). Standard errors in parentheses, clustered atperson-level. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Coefficients and standard errors multiplied by 100for presentation.

59

Table A6. Local cultural composition

Dep. Var.: Euclidean Cultural Similarity Index

(1) (2) (3)mediator baseline local-national local culturalvariable cultural distance dispersion

Panel A: Linear specification

Months since arrival 0.053∗∗ 0.052∗∗ 0.064∗∗∗

(0.022) (0.023) (0.024)

Months since arrival × mediator 0.010 -0.047∗∗∗

(0.016) (0.015)

Person-year observations 12,681 12,681 12,681Person observations 6,867 6,867 6,867R2 adjusted 0.251 0.251 0.252

Panel B: Polynomial specification (2nd order)

Months since arrival 0.151∗∗∗ 0.148∗∗∗ 0.145∗∗∗

(0.034) (0.034) (0.033)

Months since arrival × mediator 0.049∗∗ -0.099∗∗∗

(0.023) (0.025)

Person-year observations 12,681 12,681 12,681Person observations 6,867 6,867 6,867R2 adjusted 0.252 0.252 0.253

ControlsIndividual-level Yes Yes YesDistrict-level Yes Yes Yes

Fixed EffectsComposition Yes Yes YesSurvey year Yes Yes YesDistrict Yes Yes Yes

Table A6 shows baseline specifications with interaction between our main exogenousvariable, months since arrival (MSA) and other mediating variables at the local level.Column 1 reports our baseline specification. In column 2, we measure the local-national cultural distance as the standardized Euclidean distance between residentsof NUTS2-region and residents of all of Germany over the same set of questionslisted in Table A1. In column 3, we interact MSA with the standard deviation inresponses to the same set of questions by locals at the NUTS2 level. Standard errorsin parentheses, clustered at person-level. Coefficients and standard errors multipliedby 100 for presentation. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

60

Table A7. Validation of threat and mediators

Dep. Var: At least some worries about xenophobia(1) (2) (3) (4) (5) (6) (7) (8) (9)

Mediator: first threat PC share of refugees employment of shared accommodationimmigrants

Panel A: Linear specification

Months since arrival 0.076∗∗ 0.074∗∗ 0.072∗∗ 0.077∗∗ 0.053∗ 0.063∗ 0.077∗∗ 0.080∗∗ 0.081∗∗

(0.032) (0.034) (0.032) (0.032) (0.032) (0.034) (0.032) (0.035) (0.035)

mediator 3.041∗∗∗ -1.744∗∗∗ -0.085 0.480 0.484 0.415(0.584) (0.520) (0.596) (1.065) (2.041) (2.039)

Months since arrival × mediator 0.034 0.015 -0.001 -0.003 0.003(0.030) (0.030) (0.029) (0.071) (0.072)

Months since arrival × first threat PC 0.019(0.030)

Months since arrival × mediator × first threat PC 0.139(0.087)

Person-year observations 12,431 12,431 12,431 12,431 11,881 11,881 12,381 12,381 12,381Person observations 6,775 6,775 6,775 6,775 6,449 6,449 6,767 6,767 6,767R2 adjusted 0.066 0.092 0.065 0.092 0.064 0.093 0.064 0.092 0.093

Panel B: Polynomial specification (2nd order)

Months since arrival 0.055 0.107∗ 0.049 0.087 0.043 0.096∗ 0.055 0.079 0.118∗∗

(0.052) (0.055) (0.052) (0.055) (0.055) (0.058) (0.053) (0.059) (0.060)

mediator 3.035∗∗∗ -1.736∗∗∗ -0.084 0.407 -0.768 -0.749(0.584) (0.520) (0.596) (1.073) (2.581) (2.605)

Months since arrival × mediator 0.171∗∗∗ -0.076∗ -0.047 0.089 0.151(0.049) (0.045) (0.049) (0.125) (0.136)

Months since arrival × first threat PC 0.147∗∗∗

(0.054)

Months since arrival × mediator × first threat PC 0.565∗∗

(0.275)

Person-year observations 12,431 12,431 12,431 12,431 11,881 11,881 12,381 12,381 12,381Person observations 6,775 6,775 6,775 6,775 6,449 6,449 6,767 6,767 6,767R2 adjusted 0.066 0.093 0.065 0.092 0.063 0.093 0.064 0.092 0.093

ControlsIndividual-level Yes Yes Yes Yes Yes Yes Yes Yes YesDistrict-level Yes Yes Yes Yes Yes Yes Yes Yes Yes

Fixed EffectsSurvey year Yes Yes Yes Yes Yes Yes Yes Yes YesDistrict No Yes No Yes No Yes No Yes Yes

61

Appendix B. Robustness Checks

Table B1. Assimilation effort: robustness with Canberra Index

(1) (2) (3) (4) (5) (6) (7)MSA 0.022∗∗∗ 0.021∗∗∗ 0.021∗∗∗ 0.021∗∗∗ 0.021∗∗∗ 0.021∗∗∗ 0.023∗∗∗

(0.006) (0.006) (0.006) (0.006) (0.006) (0.005) (0.006)

MSAS -0.000∗∗∗ -0.000∗∗∗ -0.000∗∗∗ -0.000∗∗∗ -0.000∗∗∗ -0.000∗∗∗ -0.000∗∗∗

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Person-year observations 12,681 12,681 12,681 12,681 12,681 12,681 12,681Person observations 6,867 6,867 6,867 6,867 6,867 6,867 6,867R2 adjusted 0.157 0.157 0.177 0.182 0.209 0.216 0.237

ControlsIndividual-level No No Yes Yes Yes Yes YesDistrict-level No No No Yes Yes Yes Yes

Fixed EffectsComposition Yes Yes Yes Yes Yes Yes YesSurvey year No Yes Yes Yes Yes Yes YesFederal-State No No No No Yes No NoNUTS-2 No No No No No Yes NoDistrict No No No No No No Yes

Table B1 analyses similar to the specifications in Table 4 with the Canberra Cultural Similarity Index. Positivecoefficients indicate a reduction in cultural distance. Cultural similarity index includes: risk, reciprocity, leisureand cultural activities, satisfaction, worries, political interest, locus of control social inclusion, self-attitude, trust,egoistic-fair society. Standard errors in parentheses clustered at person-level. Regional controls are measuredat district-level in Dec-2012 (interacted with survey year dummies). All specifications control for a regressionsconstant and missing categories in covariates. Coefficients and SE multiplied by 100 for presentation. ∗ p < 0.10,∗∗ p < 0.05, ∗∗∗ p < 0.01.

62

Table B2. Probability of assignment to NUTS2 type by pre-entry characteristic

(1) (2) (3) (4) (5)Pre-entry variable gender age work exp. origin initial CSI

Panel A: High vs. Low Unemployment NUTS2 (Dec-2012)

variable * 2013, 2014 0.918 0.047 0.014 -3.872(2.466) (0.143) (0.157) (3.684)

variable * 2016 1.926 0.071 0.052 -3.258 7.534(3.072) (0.168) (0.187) (4.322) (5.446)

Question composition No No No No YesObservations 6,682 6,682 6,259 6,682 382R2 adjusted 0.01 0.01 0.01 0.00 0.04

Panel B: Urban vs. Rural District

variable * 2013, 2014 -0.036 0.034 0.021 1.457(2.264) (0.130) (0.144) (3.319)

variable * 2016 -4.419 -0.059 0.210 -1.833 1.891(2.960) (0.148) (0.173) (3.917) (5.027)

Question composition No No No No YesObservations 6,682 6,682 6,259 6,682 382R2 adjusted 0.00 0.00 0.00 0.00 0.01

Panel C: High vs. Low Threat NUTS2 (PCA)

variable * 2013, 2014 -3.475 -0.025 -0.099 -3.054(2.513) (0.141) (0.155) (3.656)

variable * 2016 0.457 0.134 0.159 1.285 1.593(3.073) (0.167) (0.184) (4.336) (5.256)

Question composition No No No No YesObservations 6,682 6,682 6,259 6,682 382R2 adjusted 0.01 0.01 0.01 0.00 0.03

Table B2 shows plausibility of identifying assumption by NUTS2 type. Year2015 (year with highest refugee influx) omitted. Panel A: Probab. of beingallocate to high vs. low unemployment NUTS2 region. Panel B: Prob. ofbeing allocated to an urban relative to a rural NUTS2-region. Classificationof NUTS2-types is based on density and distribution of population within theregion. For more information see: Federal Institute for Research on Building,Urban Affairs and Spatial Development. Panel C: Probab. of being allocate tohigh vs. low threat NUTS2 region. Estimation-Method: OLS probability model(weighted), cross-sectional, 1 observation per person. Standard errors clusteredat household level. Coefficients and SE multiplied by 100 for presentation. *p<0.10, ** p<0.05, *** p<0.01.

63

Table B3. Cultural convergence and mobility.

Dep. Var.: Euclidean Cultural Similarity Index

All Res. obl. Stayers Movers(1) (2) (3) (4)

Months since arrival 0.151∗∗∗ 0.150∗ 0.143∗∗∗ 0.180∗∗

(0.034) (0.080) (0.039) (0.087)

Person-year observations 12,681 3,867 9,546 3,135Person observations 6,867 2,884 5,267 1,729R2 adjusted 0.252 0.301 0.264 0.255

ControlsIndividual-level Yes Yes Yes YesDistrict-level Yes Yes Yes YesMSAS Yes Yes Yes Yes

Fixed EffectsComposition Yes Yes Yes YesSurvey year Yes Yes Yes YesDistrict Yes Yes Yes Yes

Table B3 shows our baseline specification by mobility. Cultural similaritymeasured as the inverse Euclidean distance to non-refugees 1) in the NUTS2region that the refugee was assigned to 2) in the NUTS2 region that therefugee was assigned to for the sub-sample of refugees falling under the res-idency obligation 3) in the NUTS2 region that the refugee was assigned tofor the sub-sample of refugees who still live in the assigned region 4) in theNUTS2 region of residence for the sub-sample of refugees who moved out-side of the assigned region. The index, control variables and fixed effects arethe same as in our preferred specification in Table 4 column 7. Coefficientsand SE multiplied by 100 for presentation. Positive coefficients indicate areduction in cultural distance. Standard errors in parentheses, clustered atperson-level. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

64

Table B4. Robustness check for out-selection from threat regions

Dep. Var.: Moved out of NUTS2-regionof assignment (dummy)

(1) (2) (3) (4)first hate crimes

threat PC against refugeesPanel A: Linear specification

MSA 0.106∗∗∗ 0.106∗∗ 0.107∗∗∗ 0.106∗∗

(0.030) (0.042) (0.030) (0.042)MSA × mediator 0.039 -0.012 0.044 -0.019

(0.032) (0.032) (0.034) (0.029)Person-year observations 12,681 12,681 12,681 12,681Person observations 6,867 6,867 6,867 6,867R2 adjusted 0.004 0.357 0.004 0.357

Panel B: Polynomial specification (2nd order)

MSA 0.192∗∗∗ 0.209∗∗∗ 0.196∗∗∗ 0.219∗∗∗

(0.041) (0.060) (0.040) (0.059)MSA × mediator 0.063 -0.036 0.094∗∗ -0.015

(0.047) (0.050) (0.047) (0.049)

Person-year observations 12,681 12,681 12,681 12,681Person observations 6,867 6,867 6,867 6,867R2 adjusted 0.004 0.358 0.005 0.358

ControlsIndividual-level No Yes No YesDistrict-level No Yes No Yes

Fixed EffectsSurvey year No Yes No YesDistrict No Yes No Yes

Table B4 estimates the likelihood that a refugee does not live in theassigned NUTS2-region at the time of the interview with a linear proba-bility model (moved = 1; stayed = 0). We interact our main exogenousvariable, months since arrival (MSA), and other mediating variables atthe local level focusing on political features of the non-refugee commu-nity in the same NUTS2 region. All specifications in Panel A includeMSA but not MSAS. Panel B also includes the interaction term betweenmonths since arrival squared (MSAS) and the mediating variable (notreported). In columns 1 and 2, we use a threat index. In columns 3 and4, we use geo-located hate crimes against refugees between 2014 and2015 from Bencek & Strasheim (2016). Coefficients and SE multipliedby 100 for presentation. Standard errors in parentheses, clustered atperson-level. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

65

Table B5. Robustness check for panel attrition based on cultural similarity

Dep. Var.: Panel attrition to subsequent survey wave (dummy)(1) (2) (3) (4) (5) (6) (7) (8)

MSA 0.138∗∗∗ 0.137∗∗∗ 0.137∗∗∗ 0.073∗ 0.075∗ 0.089∗∗ 0.091∗∗ 0.085∗∗

(0.037) (0.039) (0.039) (0.039) (0.039) (0.038) (0.038) (0.041)

Cultural similarity (z-std) 1.156 1.161 1.191 1.282 0.923 0.688 0.238(0.995) (0.995) (0.960) (0.958) (0.954) (0.956) (0.975)

MSA × Cultural similarity (z-std) 0.001 0.001 0.008 0.010 0.008 0.009 0.014(0.032) (0.032) (0.030) (0.030) (0.030) (0.030) (0.031)

Person-year observations 8,896 8,896 8,896 8,896 8,896 8,896 8,896 8,896Person observations 6,506 6,506 6,506 6,506 6,506 6,506 6,506 6,506R2 adjusted 0.017 0.017 0.017 0.049 0.051 0.060 0.062 0.077

ControlsIndividual-level No No No Yes Yes Yes Yes YesDistrict-level No No No No Yes Yes Yes Yes

Fixed EffectsComposition Yes Yes Yes Yes Yes Yes Yes YesSurvey year No No Yes Yes Yes Yes Yes YesFederal-State No No No No No Yes No NoNUTS-2 No No No No No No Yes NoDistrict No No No No No No No Yes

Table B5 analyses the probability of panel attrition to the subsequent survey wave with a linear probability model. It shows baseline specifications withinteraction between our main exogenous variable, months since arrival (MSA), and cultural similarity as a mediating variable. Columns 1 to 8 stepwiseinclude more confounding variables. Standard errors in parentheses, clustered at person-level. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Coefficients andstandard errors multiplied by 100 for presentation.

66

Table B6. Robustness check for non-response to each of the 12 questions index components

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)Risk Neg. Recipr. Pos. Recipr. Leis., cult. Satisf. (life, Worries (econ, Polit-Interest Locus Control Social Incl. Trust Pos. self att. Ego-altr.

activ, health, living) health, xeno) societyMSA -0.020 -0.015 -0.025 0.003 0.000 0.003 -0.004 -0.041 -0.003 -0.008 -0.049 0.038

(0.024) (0.041) (0.038) (0.004) (.) (0.009) (0.013) (0.042) (0.020) (0.016) (0.044) (0.052)

MSAS 0.000∗∗ 0.000 -0.000 -0.000 0.000 0.000 0.000+ 0.000+ 0.000 0.000 0.000 -0.000(0.000) (0.000) (0.000) (0.000) (.) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

MSA × AfD 2017 -0.013 0.042 0.054 -0.002 0.000 0.005 -0.009 -0.003 0.030∗ 0.029∗ -0.009 -0.059(0.021) (0.044) (0.041) (0.003) (.) (0.010) (0.012) (0.044) (0.018) (0.016) (0.045) (0.060)

MSAS × AfD 2017 0.000 -0.000 -0.000 0.000 0.000 -0.000 0.000 0.000 -0.000+ -0.000∗∗ -0.000 0.000+

(0.000) (0.000) (0.000) (0.000) (.) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)R2 adjusted 0.035 0.072 0.082 0.001 . 0.005 0.014 0.056 0.032 -0.006 0.051 0.033

MSA -0.018 -0.023 -0.033 0.003 0.000 0.004 -0.004 -0.047 -0.003 -0.007 -0.055 0.035(0.024) (0.041) (0.037) (0.004) (.) (0.010) (0.013) (0.042) (0.020) (0.016) (0.044) (0.052)

MSAS 0.000∗∗ 0.000 0.000 -0.000 0.000 0.000 0.000 0.000∗ 0.000 0.000 0.000 -0.000(0.000) (0.000) (0.000) (0.000) (.) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

MSA × NPD 2013 -0.001 0.014 0.025 -0.001 0.000 0.008 -0.009 -0.041 0.034∗ 0.023∗ -0.033 -0.105∗

(0.022) (0.047) (0.044) (0.003) (.) (0.010) (0.013) (0.048) (0.019) (0.013) (0.049) (0.054)

MSAS × NPD 2013 0.000 -0.000 -0.000 0.000 0.000 -0.000 0.000 0.000 -0.000+ -0.000∗∗ -0.000 0.001∗∗

(0.000) (0.000) (0.000) (0.000) (.) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

R2 adjusted 0.035 0.072 0.082 0.001 . 0.005 0.014 0.056 0.032 -0.007 0.051 0.033

MSA -0.017 -0.031 -0.048 0.003 0.000 0.002 -0.004 -0.043 -0.008 -0.007 -0.053 0.017(0.024) (0.039) (0.036) (0.004) (.) (0.009) (0.013) (0.040) (0.020) (0.017) (0.043) (0.052)

MSAS 0.000∗∗ 0.000 0.000 -0.000 0.000 0.000 0.000+ 0.000∗ 0.000 0.000 0.000 0.000(0.000) (0.000) (0.000) (0.000) (.) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

MSA × Number anti-refugee 0.007 0.007 -0.025 0.000 0.000 -0.001 -0.011 -0.051 0.017 0.009+ -0.076+ -0.086+

events per 100k people (0.021) (0.040) (0.042) (0.003) (.) (0.011) (0.013) (0.044) (0.019) (0.006) (0.051) (0.053)

MSAS × Number anti-refugee -0.000 -0.000 0.000 0.000 0.000 0.000 0.000 0.000 -0.000 -0.000+ 0.000 0.001∗∗

events per 100k people (0.000) (0.000) (0.000) (0.000) (.) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

R2 adjusted 0.035 0.071 0.082 0.001 . 0.005 0.014 0.056 0.031 -0.007 0.052 0.033

Individual controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRegional controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Survey year Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesDistrict Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesPerson-year observations 12,681 6,859 6,859 8,280 12,681 12,681 12,681 4,042 6,859 3,426 6,859 3,426Person observations 6,867 6,859 6,859 5,277 6,867 6,867 6,867 4,042 6,859 3,426 6,859 3,426

Table B6 analyses the probability of non-response to the survey question as denoted in the column header with a linear probability model. Outcome variable is a dummy equal to one if the individual did not answer the survey question and zero otherwise. It shows baselinespecifications with interaction between our main exogenous variable, months since arrival (MSA), and other mediating variables at the local level focusing on political features of the non-refugee community in the same NUTS2 region (z-standardized). Standard errors inparentheses, clustered at person-level. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Statistical significance levels adjusted for multiple hypotheses testing by controlling the familywise error rate (FWER) using the Romano-Wolf procedure (Clarke et al., 2019; Romano & Wolf,2016, 2005a,b). Coefficients and standard errors multiplied by 100 for presentation.

67

Table B7. Assimilation effort and success under threat - ruling out economic factors

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Base first threat PC right-wing vote (AFD 2017) right-wing vote (NPD 2013) hate crimes against refugees

Outcome Effort (CS) Success (empl) Effort (CS) Success (empl) Effort (CS) Success (empl) Effort (CS) Success (empl) Effort (CS) Success (empl)

Panel A: Linear specification

MSA 0.053∗∗ 0.495∗∗∗ 0.053∗∗ 0.496∗∗∗ 0.052∗∗ 0.497∗∗∗ 0.053∗∗ 0.495∗∗∗ 0.054∗∗ 0.495∗∗∗

(0.022) (0.044) (0.022) (0.044) (0.022) (0.044) (0.022) (0.043) (0.022) (0.044)

MSA × unemployment 0.001 -0.043 -0.018 -0.011 -0.009 -0.018 -0.016 -0.009 -0.026 -0.015(0.019) (0.033) (0.024) (0.039) (0.022) (0.036) (0.025) (0.041) (0.025) (0.038)

MSA × mediator 0.029 -0.049 0.018 -0.047 0.026 -0.051 0.040 -0.041(0.028) (0.037) (0.022) (0.034) (0.027) (0.041) (0.037) (0.036)

Person-year observations 12,681 12,681 12,681 12,681 12,681 12,681 12,681 12,681 12,681 12,681Person observations 6,867 6,867 6,867 6,867 6,867 6,867 6,867 6,867 6,867 6,867R2 adjusted 0.251 0.171 0.251 0.172 0.251 0.172 0.251 0.172 0.252 0.172

Panel B: Polynomial specification (2nd order)

MSA 0.150∗∗∗ 0.808∗∗∗ 0.171∗∗∗ 0.873∗∗∗ 0.165∗∗∗ 0.874∗∗∗ 0.166∗∗∗ 0.870∗∗∗ 0.173∗∗∗ 0.859∗∗∗

(0.034) (0.057) (0.037) (0.051) (0.037) (0.053) (0.036) (0.053) (0.036) (0.051)

MSA × unemployment 0.080∗∗ -0.108∗∗ 0.024 -0.151∗∗∗ 0.054+ -0.148∗∗∗ 0.034 -0.158∗∗∗ -0.010 -0.142∗∗

(0.032) (0.044) (0.037) (0.055) (0.034) (0.053) (0.038) (0.059) (0.040) (0.056)

MSA × mediator 0.099∗∗∗ 0.059 0.071∗∗ 0.041 0.080∗∗ 0.078+ 0.137∗∗∗ 0.057(0.036) (0.045) (0.030) (0.041) (0.036) (0.048) (0.042) (0.048)

Person-year observations 12,681 12,681 12,681 12,681 12,681 12,681 12,681 12,681 12,681 12,681Person observations 6,867 6,867 6,867 6,867 6,867 6,867 6,867 6,867 6,867 6,867R2 adjusted 0.252 0.177 0.253 0.179 0.253 0.179 0.253 0.179 0.253 0.178ControlsIndividual-level Yes Yes Yes Yes Yes Yes Yes Yes Yes YesDistrict-level Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Fixed EffectsComposition Yes No Yes No Yes No Yes No Yes NoSurvey year Yes Yes Yes Yes Yes Yes Yes Yes Yes YesDistrict Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Table B7 shows baseline specifications with interaction between our main exogenous variable, months since arrival (MSA), the unemployment rate in the assigned NUTS2 level at the timeof interview (z-standardized), and other mediating variables (also z-standardized) at the local level focusing on political and socio-psychological features of the non-refugee community inthe same NUTS2 region. All specifications in Panel A include MSA but not MSAS. Panel B also includes the interaction term between months since arrival squared (MSAS) and themediating variables (not reported). Columns 1 and 2 interact MSA with regional (baseline, Dec-2012) unemployment. In columns 3 and 4, we use the first PC threat index. In columns 5and 6, we use the NUTS2-level vote share for the anti-immigration party AfD in the year 2017. In columns 7 and 8, we take the extreme right party NPD in the year 2013. In columns 9and 10, we use geo-located hate crimes against refugees between 2014 and 2015 from Bencek & Strasheim (2016). Standard errors in parentheses, clustered at person-level. + p < 0.15, ∗

p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Coefficients and standard errors multiplied by 100 for presentation.

68