Who supports refugees? Diversity assent and pro-refugee engagement in Germany

Abstract

During the 2015 “summer of welcome”, the mass arrival of refugees to Germany triggered widely publicised acts of pro-refugee solidarity among citizens. To date, scholarship has largely focused on hostility towards immigrants—including refugees-, and few studies shed light on the determinants of acceptance, so that refugee solidarity remains poorly understood. In this paper, we hypothesise that pro-refugee engagement does not just reflect humanitarian concerns, but also a more general acceptance of socio-cultural diversity in German society. Relying on unique survey data gathered in 2019–2020, we show that a majority of urban Germans indeed engaged in some form of support. We employ latent class analysis to capture the spectrum of diversity attitudes within the urban population. A series of regression models show that diversity attitudes are powerful predictors of pro-refugee engagement. Overall, our study helps advancing immigration scholarship by demonstrating that widespread support for refugees is strongly associated with more general diversity assent.

Introduction

Autumn 2015, with Germans cheering at train stations to welcome refugees and many engaging in further support, left many political and scholarly observers puzzled. Indeed, in the past two decades, scholarly attention has been directed more intensely at describing and explaining anti-immigrant sentiment and behaviour rather than pro-immigration and pro-diversity stances, about which little has hitherto been known.¹ As negative responses to immigration have been often associated with extreme Right tendencies, understanding them has seemed particularly urgent. Yet, analysing the forms and motivations of pro-immigrant and pro-diversity stances is arguably no less relevant. In Germany and other European countries, they represent majority views (Dennison & Geddes, 2019). Further, in order to understand the political dynamics in a country, we need to comprehend both dissent and assent vis-à-vis immigration and migration-based diversity. The surprise engendered by the reception of refugees in 2015 suggests that there are important gaps in our scholarly understanding of such patterns of transnational support. Although the extent of solidarity with refugees witnessed in and beyond autumn 2015 has stimulated new research, we still insufficiently understand the wider significance of refugee support and what undergirds it. Who are those that engaged in support of refugees during and after 2015, and how does refugee support relate to broader attitudes vis-à-vis immigration and diversity in Germany? To date, there exist very few larger, representative studies of support towards refugees and its associated factors among ordinary Germans.

This paper aims to help alleviate this unsatisfactory situation by pointing to the significance of more general attitudes to diversity as linked with support for refugees in Germany. Based on unique survey data gathered in 2019–2020, we examine the scope and structure of refugee support in the general urban population. By demonstrating wide-ranging support for refugees and its links with more general patterns of assent to societal diversity, our results contradict assumptions according to which support for refugees was just a momentary phenomenon and an eruption of little politicized humanitarian feelings (see e.g. Fassin, 2016; Kleres, 2017: 138). Instead, we argue that refugee support is indicative of a more general disposition entailing a positive evaluation of an increasing variety of lifestyles and of internationalized, more varied environments. The accepting and approving views towards diversity as a feature of urban life in Germany may be closely associated with favourable responses to new migrations and solidarity with foreigners. The rest of the paper is organized as follows: we discuss the state of research and outline some theoretical considerations on pro-diversity attitudes. Next, we discuss the data our analyses are based on, and present our empirical results. A fifth section offers conclusions and summarizes our argument.

State of research and theoretical considerations

Scholars and other observers widely agree that Europe experienced extraordinary events in summer 2015. Not only was the mass movement towards the center of Europe unparalleled in the decades following the Second World War and its aftermath (Ataç et al., 2016: 528), the willingness of hundreds of thousands of Europeans to help these refugees represented “an unprecedented solidarity wave in Europe” (Rea et al., 2019: 25). While similar developments occurred in several countries, German developments were perceived as particularly striking, as despite large refugee arrivals, a surprisingly high number of citizens were willing to donate and help. As German Green Party politician Göring-Eckart (2015: 11614) proudly claimed, Germans had become the world champions of solidarity.

Past work on attitudes towards refugees

The 2015 summer of migration and solidarity has instigated much scholarly activity. And yet, there remain relevant gaps in sscholarly knowledge, most obviously regarding a better understanding of the motivations underlying contrasting positions and mobilisations around refugee reception (see also Koos & Seibel, 2019: 705). Having reviewed recent political science contributions, Dražanová (2022: 85) notes: “Disagreement over what drives people’s attitudes to immigration persists.” Similarly, a recent cross-national meta-analysis into factors determining prejudice towards refugees provides some insights, but no conclusive answer (Mei Cowling et al., 2019: 513).

Given the large body of research that has examined attitudes to immigration (and to a lesser extent diversity), this may seem surprising. Based on the European Social Survey (ESS), a recent collection of articles investigated what drives attitudes towards immigration in European countries and compared differing patterns across countries (Heath et al., 2020; see also Hainmueller & Hopkins, 2014). As in that collection, one focus of research has been on how (ascribed) characteristics of the immigrants themselves influence responses to them. Studies show that those fleeing persecution are more positively assessed compared to other immigrant groups (Abdelaaty & Steele, 2020; Bansak et al., 2016; Czymara & Schmidt-Catran, 2017; Hager & Veit, 2019; Meidert & Rapp, 2019). In sum, it has been well-established that Europeans currently tend to respond more positively to immigrants if labelled “refugees” or if attributed additional characteristics stressing their deservingness of help. Indeed, when in 2015–2016 hundreds of thousands of migrants arrived in Europe, leading politicians and a majority of media outlets in Germany suggested that they were refugees fleeing from persecution and war and legitimately seeking help (Haller, 2017; Maurer et al., 2021). In all likelihood, this framing provided a basis and legitimacy for citizen engagement.

Past work gauging and explaining pro-refugee engagement

So far, the most reliable source among efforts to assess the scale of active engagement in support of refugees in Germany is the “Barometer of Public Opinion on Refugees in Germany” conducted in 2016, in addition to the regular Socio-Economic Panel (Jacobsen et al., 2017). Run monthly for a year, the survey provides representative results on engagement levels and attitudes to refugee reception and its implications—however only for 2016. Around one third of Germans had, in the course of the year preceding the interview, supported refugees by donating money and other things or by helping them in other ways (Jacobsen et al., 2017). The regular German Survey on Volunteering now includes questions on engagement in support of migrants and refugees. The focus here is more narrowly on somewhat consistent forms of engagement in associations or projects. 12.4% stated that between 2014 and 2019 they had engaged in this way for asylum-seekers or refugees (Kausmann et al., 2022: 213–214).

While there exists a large body of research both on general motivations of voluntary engagement and on social movement activism (e.g. Arriagada & Karnick, 2021; Wilson, 2000), research focused on refugee support expanded only recently. With its multiple forms, it should be distinguished from a social movement (Roth, 2021). Further, supportive behaviour in a wider sense, including donations and spontaneous help, is not identical with volunteer engagement in consistent structures. Following the surprising upsurge of pro-refugee engagement, several studies have investigated characteristics of the engaged individuals and their motivations. One much cited German study by Karakayali et al., (2016: 11–19) surveyed activists in refugee support groups, first in 2014, and again at the end of the year 2015. With more than 2000 participants in 2015, the sample is large.² About two thirds of the 2015 supporters were new to the field and had previously not actively supported refugees (for similar observations in the UK and France, see Maestri & Monforte, 2020). As the authors point out, with the growth of volunteer engagement, the social-demographic profile of the supporters became more similar to that of the population overall, although the refugee-supporters were on average more likely to be female and urban (Karakayali et al., 2016: 3). A range of political positions was now represented (Karakayali, 2017: 23), such that pro-refugee engagement ‘has activated individuals from the socio-political centre of society, well beyond the previously committed radical-left, antiracist, and faith-based groups” (Fleischmann & Steinhilper, 2017: 17).

The extension of volunteer engagement and support of refugees to broader social strata and political commitments has thus been identified as one key feature of the so-called “summer of refugees”. Still, if the supporters are a cross-section of the population in socio-demographic terms and encompass people with differing political views, we remain wondering what motivates supportive behaviour. Evidence on the relevance of religious commitment for activists in Germany remains inconclusive (Nagel & El Menouar, 2019).³ Nevertheless, scholars broadly agree that some form of general humanitarian commitment seems to underlie engagement in support of refugees and to provide the “smallest common denominator” (Karakayali, 2017: 19). Kumbruck et al., (2020: 69, 76–77) identified a willingness to help as more or less universal motivation among engaged individuals in Germany. In laboratory experiments, Böhm et al. (2018: 7287) found that a “pro-social concern towards others” provided a strong motivation to help refugees, even at a cost to oneself. Before 2015, scholars observed that “humanitarianism appears to be a common and powerful motivating force among those taking action on behalf of immigrants” (Newman et al., 2015: 584). Thus, Sandri (2018) has analyzed ‘volunteer humanitarianism’ around refugee camps in Calais. Volunteers supporting migrants across the Mexican-US border have been described as motivated by different kinds of humanitarian stances (Gomez et al., 2020). Still, does reference to widely shared humanitarian commitments suffice to explain the wide-ranging engagement in support of refugees witnessed since 2015? And what exactly does a humanitarian commitment entail? Some scholars suggest that it lacks a political edge and is often uncritical of predominant restrictive refugee policies (e.g. Kleres, 2017: 138), while Gomez et al. (2020) conversely offer a typology of humanitarian motivations that includes political activism as well as moral and religious motives to “do good”.

The case for diversity attitudes as predictors of refugee support among the urban population in Germany

Beyond traditional humanitarian convictions, we hypothesize that the positive reception of refugees in and beyond 2015 is associated with broader attitudes vis-à-vis diversity in contemporary German society. While a desire for ethnic homogeneity has historically been strong in Germany (e.g. Brubaker, 1992), socio-cultural heterogeneity is nowadays widely accepted if not valued (Schönwälder et al., 2016). Such positive diversity attitudes can be considered part of broader and longer-term cultural transformations towards a higher regard for individual and minority rights as described by Norris and Inglehart (2019) see also Inglehart, 2018) in their theory of economic growth and cultural change. In postindustrial economies such as Germany, large progress in material welfare and existential security have shifted sizeable portions of public opinion towards “postmaterialist values”, which includes the increasing acceptance of sexual, ethnic, and other minorities (Inglehart, 2018: 25–360). Postindustrial cities have been harbingers of such change: in Germany and other European countries, larger cities tend to be more pro-immigration compared to other regions, in large part because they concentrate populations with higher educational attainments and higher-paying jobs (Maxwell 2019). In sum, we reason that refugee support is strongly related to broader pro-diversity attitudes that have blossomed among the German urban population in recent decades (Schönwälder et al., 2016).

Recent studies include evidence pointing towards this interpretation of refugee support. Thus, Karakayali et al. (2016: 32–33) mention the importance of cosmopolitan and antiracist attitudes among volunteers: More than 90% of those active in 2015 and surveyed by them agreed that they wanted to learn new things about the world and other cultures through their engagement and that they aimed to set a sign against racism. Kumbruck et al., (2020: 75–79) similarly report that interview partners stressed how much they learned through their engagement, that it was like “travelling the world” (see also Fleischmann, 2020: 238; Perron, 2020). Such citations resemble statements recorded in an earlier large-scale study of diversity experiences in German urban neighbourhoods. Based on evidence from 50 neighbourhoods in mid- and large-sized cities, the “Diversity and Contact” study found that diversity has become normalized in perceptions of urban life and, as such, widely acknowledged and even appreciated (Schönwälder et al., 2016).

Studies on the foundations of solidarity point in a similar direction (Bauder & Juffs, 2020). As Nowicka et al. (2019: 387) assert, a new phenomenon of transnational solidarity has emerged whose “normative underpinnings” are distinct from those of solidarity based on common nationality. Indeed, the widespread solidarity with refugees often coming from other continents seems to contradict assumptions that solidarity is restricted to nation-state contexts. Several scholars argue that new forms of interaction and mutual support develop in social contexts marked by socio-demographic diversity (Oosterlynck et al., 2016; Portes & Vickstrom, 2011; Schönwälder et al., 2016). The arrival of a large number of refugees may have provided an opportunity to act on such widely shared pro-diversity beliefs. However, we acknowledge that the relationship between diversity beliefs and refugee support may be more dynamic and co-constitutive, particular under the effect of refugee-native contact and the diffusion of pro-refugee attitudes in the general population following the mass arrivals of 2015. Hence, we do not assume that an association between diversity attitudes and refugee support only flows from one to the other. Rather, our overall goal in this paper is to study the association between diversity attitudes and refugee support, thus providing an important first step towards establishing the analytical importance of diversity attitudes for broader scholarly understandings of intergroup relations in migration societies like Germany.

Data and empirical strategy

In this paper, we exploit a unique dataset on support for societal diversity among the general population in Germany—the Diversity Assent (DivA) survey. The survey instrument was designed to fill a specific gap: while much past research focuses on understanding determinants of hostility towards minority groups, there is a dearth of research and data on what motivates those who oppose right-wing positions. This survey focuses on measuring the experience and perception of diversity as well as attitudes towards and support for specific groups, including refugees. Our focus here is predominantly on pro-diversity attitudes, which we wished to describe in their nuance and heterogeneity. To do so, we geared our research design towards medium- and large-sized German cities. To be clear, our research design does not assume refugee support did not occur in rural parts of Germany—it did, although Karakyalali et al. (2016: 3) suggests most refugee supporting activities occurred in mid- and large-sized cities. Rather, we reason that cities concentrate more pro-diversity attitudes (Maxwell 2019) as well as a wider range of diversity attitudes in general due to their size and their inherent heterogeneity (Fischer, 1975). Hence, they offer an appropriate social setting to study the relationship between diversity attitudes and refugee supporting activities.

The survey was administered by telephone between November 2019 and April 2020 on a random sample of 2,917 respondents living in twenty randomly selected German cities, through a dual-frame strategy mixing landlines and mobile numbers. The survey had a response rate of 5.6%—which is low in absolute terms, but in line with rapidly declining response rates to telephone surveys in general (Berinsky, 2017; Couper, 2017; Keeter et al., 2017). Research based in the United States, where issues of nonresponse in telephone surveys have appeared earlier than in Europe, has shown that low response rates need not be conflated with response quality or nonresponse bias (Keeter et al., 2017), particularly if high quality auxiliary data are available for post-survey calibration (e.g. Groves, 2006; Keeter et al., 2006; Koch & Blohm, 2016). This is our case here. In the descriptive analyses that follow, we calibrate our estimates with poststratification weights using rich data from the Mikrozensus, the largest household survey conducted by federal and state statistical offices in Germany. Specifically, we use the Mikrozensus to construct reference points on high-dimensionality cells for multiple sociodemographic variables of interest (municipality size, age, education level, and gender), adjust for over- or underrepresentation on these cells, and accordingly weigh results from our empirical analyses. More information on poststratification weights, as well as supplementary results from weighted regressions are available in Additional file 1: Appendix C. Full technical details on the survey are available in a dedicated report (Drouhot et al. 2021). Weighted results are representative of adults living in cities of at least 50.000 inhabitants.

As in most surveys, social desirability bias is of potential concern. Indeed, our survey instruments taps into socially sensitive topics for which respondents may feel pressured to answer in socially acceptable ways, and there exists no authoritative estimate of pro-refugee engagement to judge whether respondents over- or underreported their activities. However, we do not think it threatens the validity of our findings here for two reasons. First, existing methodological work experimentally comparing self-reports on donations (a dimension of refugee support in this study) in Germany suggests such bias is small if not negligible (Gricevic et al., 2020). Secondly, we mainly wish to precisely estimate the relationship between refugee support and what we assume are its predictors in our regression models. Hence, the key issue is whether an assumed level of bias differs between support and our predictors (particularly diversity attitudes) so that our parameter estimates would be impacted. We have no reason to assume any difference in measurement error between refugee support and its predictors, and hence regard our parameter estimates as credible.

Empirically, we first describe refugee support and then move on to analyse its associated factors. As we are particularly interested in the significance of attitudes to diversity, our empirical strategy is articulated in two distinct steps. First, we use multiple survey questions on perceptions and attitudes towards diversity to uncover distinct configurations of diversity attitudes. We do so by applying latent class analysis (LCA), a model-based, inductive data partitioning approach allowing researchers to identify distinct groupings in data and falling under the general umbrella of unsupervised machine learning (Molina & Garip, 2019). It is analogous in spirit to popular clustering approaches such as k-means clustering, but has the noted advantage over the latter of accommodating multiple answer formats and types of variables (e.g. categorical, nominal, etc.), as well as being probabilistic rather than deterministic in the identification of groups (Vermunt & Magidson, 2002). It also yields discrete groupings, unlike factor or principal component analysis. This is important to our approach here, because we wish to identify qualitatively distinct groups and describe the heterogeneity of pro-diversity attitudes rather than subsume them under a continuum erasing within-group differences among those who hold prodiversity views.⁴ LCA estimates a latent (unobserved) variable accounting for the covariance between the (observed) attributes used to construct groups. This variable is assumed to have a categorical distribution with each value corresponding to a “latent class” (group) in the data. LCA produces posterior probabilities which indicates the degree of membership of observations in each unobserved class on the basis of their observed attributes. Following the assignment strategy implemented by Drouhot and Garip (2021), we use these probabilities to create multinomial distributions and classify observations at random, effectively resulting in a probabilistic assignment of observations to classes. We do so in a two-way approach, whereby we start with a broad 3-group solution, of which we further disaggregate a large middle group, yielding five subgroups in total.

In a second step, we use our obtained subgroups measuring typical configurations of diversity attitudes as explanatory factors of refugee support. Given that our dataset is cross-sectional, we of course do not claim diversity attitudes to have a causal relationship with refugee support, although such a relationship is plausible. Rather, we employ terms like “explanation” and “effect” to refer to correlative relationships only. Specifically, we use the configurations of diversity attitudes uncovered in our analysis as predictors of refugee support across multiple regression models, and strive to gauge their explanatory power vis-à-vis other potential predictors of refugee support (e.g. humanitarian orientation, age, education). Our empirical analyses proceed as follows: first, we describe the extent and structure of refugee support documented among respondents—the outcome variables we wish to shed light on in our study. We then show our configurations of diversity attitudes derived from the LCA, as well as some relevant technical details used in obtaining it. Finally, we explore statistical relationships between diversity attitudes and refugee support through regression models.

Empirical analyses

Refugee support in urban Germany—structure and intensity

We measure refugee support by asking respondents if they have been “active in the last 4 or 5 years, i.e. since 2015, with regard to refugees”. Multiple answers were possible: “by helping refugees yourself”, “by donating”, “by taking part in a demonstration to support refugees”, as well as “by taking part in a demonstration against refugees”.⁵ Only 1% of respondents report the latter. Hence, opposition to refugees is largely unheard of in our data, and we focus on describing support for refugees in all analyses that follow.

Figure 1 (top) describes the relative frequency of each activity and combination of activities in our data. Overall, weighted tabulations indicate that a relatively large portion– well over a third—do not report any refugee supporting activity. Conversely, approximately 62% of the urban population studied here engaged in some form of refugee support. We believe that our focus on the urban population and the long timespan of five year of supportive engagement account for the high number and the difference in comparison with the findings of the DIW Barometer for 2015–2016 (Jacobsen et al., 2017, see above). 46% of our respondents donated in favour of refugees, 41% engaged in direct help, and 14% demonstrated in favour of refugees. Donating emerges as the smallest common denominator of refugee support: most of those who helped or demonstrated also donated. Approximately 18% report they both helped and donated, constituting the most commonly occurring type of multi-activity support. Finally, a relatively small but nevertheless noticeable core of intense supporters (7%) had since 2015 engaged in all three ways. As Fig. 1 (bottom) further shows, the most frequently occurring level of support is one type of activity, which represents approximately 31% of respondents in our data. 24% had engaged in two activities.

Fig. 1

Structure (top) and intensity (bottom) of refugee support in urban Germany. Note: All quantities are weighted for survey design and poststratification

While we cannot tell when in the five years preceding the survey our respondents donated, helped, or demonstrated in support of refugees, our survey includes a more direct measure of the extent to which Germans in 2019–2020 favoured support for refugees. We asked survey participants whether, in Germany, enough, too little or too much was being done for refugees. Clear majorities were in favour of providing for the specific needs of refugees, only 10% wanted less support. In contrast, more than one third wanted to see support extended. This altogether suggests that support for refugees remained popular beyond 2015–2016.

Configurations of diversity attitudes

In this study, we wish to examine the relationship between diversity attitudes and variation in support for refugees. To do so, we first identify configurations of diversity attitudes in our sample through latent class analysis on items measuring respondents' level of agreement to several questions from the  DivA survey. The set of questions includes general views of diversity, more demanding views on redistribution and representation as well as opinions on the rights and support for specific minorities. We partly use answer formats that provide a stricter test of pro-diversity stances than others. We include answers to the following ten questions:

• Whether young people benefit from contact with peers of different origins and faith (agreement scale).

• Whether it is a good thing that many languages can be heard in the streets (agreement scale).

• Whether public funding for cultural activities should include the cultural traditions of minorities (agreement scale).

• Whether parliaments should reflect the diversity of the population through their members (agreement scale).

• Whether Muslims in Germany should have the right to build mosques, including in the respondents’ own neighbourhood (agreement scale).

• Whether all religions should be taught in school equally (as opposed to a focus on Christianity in school lessons—forced choice format).

• Whether the share of disadvantaged groups in the public service should be increased (as opposed to recruitment being all about suitability and competence—forced choice format).

• Whether the media should report more on cases of discrimination (as opposed to less—forced choice format).

• Whether too little is being done to meet the specific needs of Muslims in Germany (as opposed to too much or enough being done—multiple choice format).

• Whether too little is being done to meet the specific needs of gays and lesbians in Germany (as opposed to too much or enough being done—multiple choice format).

Using latent class modelling on these survey items allows us to summarize them in a parsimonious way—through a single set of categories (henceforth “diversity configurations”) which we can then use to study variation in refugee support, as opposed to using ten different questions as predictors of refugee support.⁶

Researchers using LCA often employ fit indices (such as the Bayesian Information Criterion) to determine the number of classes best fitting the data at hand. In our case, initial analyses indicated that a 5-class solution was best, but yielded ambiguous results in practice—such as groups that were redundant or hardly interpretable. Hence, and in line with past work on the primary importance of the meaningfulness and interpretability of clustering solutions (Grimmer & King, 2011), we proceeded in a two-step approach to classification, by first selecting a three-group solution. This initial 3-group solution yielded one large, diversity assenting group, one smaller but nevertheless sizeable of individuals with negative views, and one large, intermediate group with moderately positive attitudes. In order to gain further granularity, we disaggregate each larger group into two smaller ones, using another round of LCA with the same probabilistic assignment procedure outlined earlier. Our final classification solution is thus a 5-group one—as originally indicated by the fit indices—obtained through two round of LC modelling. Additional file 1: Appendix A provides further technical details for the interested reader. Table 1 shows the classification results obtained through this two-step approach. Two large groups are further disaggregated into two subgroups each, yielding five groups of diversity attitudes in total, which we now describe.

Table 1 Configurations of attitudes towards various aspects of diversity in urban Germany

Overall, we identify five subgroups ranging from strongly pro-diversity to rather sceptical. On the left-hand side of Table 1, we clearly identify a group of respondents that supports diversity without reserve: that is, an overwhelming share of these respondents supports the construction of mosques, views intergroup contact as valuable, thinks parliaments should be more reflective of the diversity of the population and that religion lessons in schools should also be more inclusive of religions other than Christianity, among other measures. 40% agree to an increase of disadvantaged groups in recruitment to the public service—the highest share on what is by far the most contentious question in our survey instrument. What is most characteristic of this first group is also a high level of assent to supporting the needs of specific minorities, such as Muslims and gays and lesbians. We label this group the “committed”.

In contrast to the “committed”, a second set of respondents do not respond as positively to questions regarding interventions to accommodate the needs of specific groups—with shares calling for more such measures that are only a quarter and a third of what they are for the committed. Hence, we label this group of diversity supporters the “hands-off”. It is plausible that this subgroup differs from the committed in how it values state interventions and public support of specific groups.⁷ Except these latter questions, members of this second group are generally supportive of diversity, albeit in magnitude always smaller than the committed.

The third group is typically, albeit moderately, positive in its outlook, altogether expressing some degree of diversity support, but at a somewhat lower level than the “hands-off” group. Like the latter, this group does altogether not favour more intervention for the needs of Muslims or gays and lesbians. We label this group “moderate” supporters.

A fourth, relatively large group is characterized by reluctance to support diversity as it concerns Muslim minorities, language plurality and public funding for minorities. It shows support, albeit less than other configurations, for the benefits of exchange among young people, even for diverse parliaments. More importantly, and not shown on the table, it features many “neither agree nor disagree” answers for questions with an agreement scale. These are more typical than negative answers. Thus, for instance, on public funding for minorities, 55% do not offer an opinion while only 5% oppose it. We label this group the “undecided”.

Finally, we identify a sizeable subset of respondents with rather negative attitudes across the board, even for general questions where positive answers do not entail a strong stance, such as the value of intergroup contact. We label this group the “sceptical”.

At the bottom of Table 1, weighted shares show the relative importance of each subgroup within the urban population in Germany (those living in cities larger than 50,000 in population). Approximately 28% have a clearly positive view of diversity—specifically, 9% form the “committed”, while a larger share of 19% are the “hands-off”. An additional 19% are moderate supporters of many aspects of diversity and in favour of representing it in parliaments, funding policies and school education. The use of weights reveals that the fourth group, the “undecided”, are larger in the population compared to what their raw frequency in the sample suggests, accounting for approximately 30%. They appear as a large, intermediate group between the pro-diversity pole, on one hand, and those who are “sceptical” on the other hand. The latter account for around 23% of the urban population. Overall, these results suggest considerable heterogeneity among pro-diversity urban populations in Germany. Our results uncover large subgroups that are moderately pro-diversity, such as the hands-off, the moderate, and to a lesser extent the undecided—together forming a numeric majority about which scholars know very little. Altogether, this degree of complexity vindicates our choice to use LCA, although we acknowledge that the subgroups form a logical continuum.⁸

Diversity attitudes as predictors of support for refugees

In a third step, we want to study how the propensity to engage in refugee support and diversity attitudes relate to one another. To do so, it is useful to start with inspecting the association between the intensity of refugee support and diversity attitudes, shown in Fig. 2.

Fig. 2

Intensity of support for refugees (number of activities) and attitudes towards diversity. Note: All quantities are weighted for survey design and poststratification

Figure 2 suggests a clear and systematic association between the intensity of refugee support and diversity attitudes, whereby the relative proportion of those who did not engage in any support increases from left to right—that is, towards configurations of diversity attitudes that are more hostile. However, one notes that refugee support remains rather frequent even among those holding more cautious attitudes towards diversity: approximately 46 and 68% engaged in some support among the sceptical and the undecided, respectively.

To further understand how diversity attitudes relate to refugee support, we estimate a series of regression models including control variables measuring known predictors of refugee support (e.g. humanitarian orientation, see above) as well as other common predictors of attitudes towards immigrants and other minority groups (e.g. education, gender, voting intention; Hainmueller & Hopkins, 2014). To ease interpretation, we binarize all continuous predictors, and graphically report all results as plotted coefficients and predicted probabilities for latter models (Kastellec & Leoni, 2007). We create a binary variable to capture missing income so as not to lose the relatively large part (18.56%) of the sample who did not report income, but otherwise do not adjust or correct of missingness on other variables since it is (fortunately) rather rare in our data. The smallest analytical sample across our models (which we allow to vary slightly so as to maximize statistical power in each analysis) captures 96.4% of our total sample. While there exists a consensus on the necessity to use weights to report means and proportions in descriptive analyses, social scientists remain divided on the importance and necessity of weights in the context of regression modelling (Bollen et al., 2016). In the results that follow, we report unweighted results due to their higher statistical power, but we refer the interested reader to Additional file 1: Appendix C for weighted models, which feature generally similar but more conservative estimates (i.e. with larger standard errors).

We start by analyzing the number of refugee-supporting activities respondents engaged in within the five years prior to the survey via ordinary least square regression (OLS). Given its distribution as seen in Fig. 1 above, we also used a poisson regression to model the underlying (count) data and arrived a substantively identical results—we choose to present results from the OLS due to them being more readily interpretable and intuitive.⁹ We model the probability for respondents to engage in any refugee supporting activity via logistic regression. Figure 3 shows the results.

Fig. 3

Factors influencing support for refugees. Note: This figure shows results from an ordinary least square regression for the number of refugee-supporting activities engaged in past 5 years (left) and a logistic regression model of the probability to engage in any refugee support activity in past 5 years (right, coefficients express change in odds ratios). Error bars are 95% confidence intervals from robust standard errors. Scales vary across models to preserve visibility

Both models show broadly consistent patterns in determinants of refugee support in our sample. In both cases, female, highly educated and high-income respondents are more likely to engage in support. Independent of these aspects, religiosity,¹⁰ having a strong humanitarian orientation¹¹ and a specific voting behaviour are all predictive of refugee support in the expected direction—i.e. the highly religious and committed humanitarians are more likely to have engaged in any support. Conservative (CDU or CSU) voters are less likely than social democratic voters to have actively supported refugees. As expected, extreme Right (AfD) voters appear far less likely to report refugee supporting activities, but we do not include this parameter estimates in the visualization since the large confidence intervals (due to their making up only 2.85% of our overall sample) results in poorly visible graphs.

Holding constant socio-demographic and common attitudinal predictors, one can readily see that diversity attitudes are strongly and consistently related to refugee support. Compared to the undecided, both the hands-off and the committed supporters are far more likely to have engaged in any support. They also tend to have engaged in multiple supporting activities, while the sceptical comparatively engaged in less. The magnitude of the difference between the five configurations of diversity attitudes is larger than the effect of any other predictor in our models. For instance, a committed supporter has an odds ratio of approximately 2.52 of having engaged in any supportive activity compared to the undecided, meaning they are approximately 2.5 times more likely to report having engaged in refugee support compared to the undecided, and net of all other factors included in our models. Although the confidence intervals around that estimate are large, this effect is substantial in magnitude.

Next, to study how different types of support might have different predictors, we model the probability of engaging in the three different types of activity through separate logistic regression models. Figure 4 shows the results of all three models simultaneously.¹²

Fig. 4

Logistic regression models for distinct refugee supporting activities. Note: Coefficients express change in odds ratios. Error bars are 95% confidence intervals from robust standard errors. Scales vary across models to preserve visibility

Helping and demonstrating show a rather similar picture; motivations and characteristics of the people engaging in such activities are apparently similar. Donating seems to be a different kind of activity. Donating is more typical for older, wealthier respondents and for women. Those engaging in the other two activities, that is, those who personally helped refugees and demonstrated to support them, are distinct in some dimensions. Those helping tend to be more religious. Those demonstrating are more often strongly humanitarian and less likely to vote for a conservative party. Altogether, the effect of specific voting preferences is relatively weak and inconsistent, although declaring a preference for the extreme right AfD has a large (albeit imprecisely estimated) negative effect on helping and donating, which we do not visually report. Most importantly, committed and hands-off diversity supporters are, all else equal more likely to help and demonstrate than the undecided—the baseline category in our models.

In contrast, the socio-demographic predictors tested here do not explain helping and demonstrating very well. For these activities, it does not matter much whether people live in a smaller or larger city, whether they are older or younger. However, and somewhat unexpectedly, respondents with a migration background are, all else equal, less likely to have donated or demonstrated in favour of refugees (but as likely to have helped). To an extent, the assumption in the literature that pro-refugee engagement engages people of different socio-demographic backgrounds is thus true. Further, results show that diversity attitudes consistently predict refugee support across its various dimensions. “Committed” diversity supporters are respectively 2.76 and 4 times more likely to have helped and demonstrated compared to the “undecided”, for instance. Even interpreted conservatively—on the lower end of the large range of values within the 95% confidence intervals—the magnitude of the difference between various types of diversity attitudes and the undecided—the baseline—dwarfs the effects of any other variable included in the models. Alternative specifications coding latent class membership as a continuous variable yield substantively identical results, with the estimate for ordered membership (from least to most pro-diversity) also dwarfing other variables in the model. This difference in magnitude is clear insofar as all other continuous variables are dummy coded, which maximizes their effect. These results are available in Additional file 1: Appendix D.

This is not as much the case for donating toward refugees, even though diversity attitudes remain sizeable in magnitude.

Finally, to illustrate the relative strength of diversity attitudes in relation to refugee support, we computed predicted probabilities pitting together diversity attitudes against other factors in our model (see Additional file 1: Appendix B). Substantively, they show that the difference in the propensity to have engaged in refugee support between those holding opposite diversity attitudes is comparable in magnitude to the cumulative weight of multiple socio-demographic and attitudinal predictors. This includes age, income, educational attainment, religiosity and humanitarian orientation.

Conclusion

What led so many ordinary Germans to engage in pro-refugee activities during autumn 2015 and beyond? This has been and continues to be a puzzle to many observers. This paper has aimed to shed light on the phenomenon and its associated factors within the urban population. Focusing on Germany, we hypothesized that this phenomenon may be closely associated with more general perspectives towards a diverse society. By and large, our empirical results based on unique survey data bear this out, as we have documented a substantial and systematic association between diversity attitudes and pro-refugee activity. More than sixty per cent of the urban population studied here engaged in some form of refugee support between 2015 and 2020. Donations were the preferred form, but many also actively helped or even participated in demonstrations. Our key contribution is thus to demonstrate that support for refugees in urban Germany has a basis not only in general humanitarian commitments, but also in a broad acceptance of diversity—which is itself diverse in its attitudinal structure. Going forward, future research ought to shed light on population segments that are broadly pro-diversity albeit with some reservations, and which we identified in our LCA results. Doing so will be important as they represent a numeric majority, yet one about which scholars still know very little.

Three important reservations should be noted: First, our survey was conducted in urban contexts, and a detailed investigation of potential differences between urban and rural publics remains wanting. Cities are appropriate settings to study the linkages between diversity attitudes and refugee support, as city-dwellers generally hold more positive attitudes than their rural counterparts (Maxwell 2019) and because cities offer much higher heterogeneity in attitudes more generally (Fischer, 1975). While we are not aware of evidence pointing in that direction, theoretically, motivations for supporting refugees may differ across regions. Investigating refugee support in rural settings including potentially specific motivations is a research lacuna. Second, we cannot tell whether our findings travel beyond Germany. Referring to immigration, Heath et al., (2020: 9) have recently emphasized the variety of European publics and pointed at potentially different drivers of anti-immigration sentiment. For instance, the German case was characterized by a high degree of elite consensus on welcoming refugees. Given past research highlighting the role of elite discourses in activating latent policy preferences (e.g. Czymara, 2019) this may well have contributed to the broad scope of support in German society. More comparative research to isolate factors associated with widespread support for refugees or its absence would be desirable. Third, the cross-sectional nature of our research design prevents us from investigating issues of causality further. Hence, our survey data and the retrospective approach we took in the survey design do not allow us to say that diversity acceptance preceded refugee support. Diversity acceptance may well have grown in consequence of positive experiences in supporting refugees or in response to related societal debates. And yet, as our findings echo research on pro-diversity attitudes in urban settings in Germany based on surveys fielded well before the mass arrival of refugees in 2015 (Schönwälder et al., 2016), we think it is likely that pro-diversity attitudes were already widespread at that time. Nevertheless, the nature of the causal relationship between diversity attitudes and supportive engagement towards minority groups (like refugees in our case) should be the object of future research.

Overall, our study suggests that support for refugees was not a short-lived expression of empathy, and may instead have specific foundations beyond humanitarian attitudes. Germany’s urban population is nowadays marked by widely held positive views of socio-cultural diversity. Most see diversity as positive for society, and often support measures to ensure equal treatment. Active support for refugees can be interpreted as expressing the wish to act on such beliefs—in different forms and intensity, again reflecting heterogeneous attitudes to diversity.