Which groups are mostly responsible for problems in your neighbourhood?

Abstract Why and under which conditions do people employ ethnic categories rather than others (such as age, class, gender, and so on) to conceptually organize their social environment? This article analyses an open-ended question on who is seen as responsible for neighbourhood problems taken from a recently conducted large-scale survey in Germany. By doing so, this study tries to give novel insight on native Germans' use of ethnic folk classifications and aims to identify contextual factors that might explain why people characterize problem-groups in ethnic terms. This article shows that drunkards, the elderly and especially teenagers are seen as problem-groups more frequently than any ethnic minority. Conditions of economic decline and out-group size are analysed as to whether they are associated with a higher likelihood of using ethnic categories. The findings suggest that the effects of out-group size are diminishing in their impact, whereas the effects of economic decline are accumulating in strength.


Introduction
The phenomenon of ethnic diversity has grown in importance in Western countries over the last decades. Some scholars highlight challenges that are associated with an ethnic diversification of the population, by showing that ethnically more diverse regions are characterized by, among other things, lower levels of public goods production (e.g. Alesina, Baqir and Easterly 1999) and lower levels of Ethnic and Racial Studies Vol. 36 No.  support for welfare state spending (e.g. Eger 2010). Given these findings, one is inclined to start hypothesizing why different ethnic groups struggle to share and produce public resources together. Such attempts are criticized by scholars who see ethnicity as cognition and object that ethnic categories do not necessarily reflect real existing groups which might struggle with one another, but are first and foremost cognitive distinctions that actors impose on the world: 'They are ways of recognizing, identifying, and classifying other people' (Brubaker, Loveman and Stamatov 2004, p. 47). While this position qualifies over-simplistic views on ethnicity by emphasizing the key role of actors' cognition, it does not see ethnicity as a purely subjective and hence arbitrary perception; ethnic categories may overlap with cultural differences as well as clusters of networks, but this does not necessarily have to be the case (Wimmer 2008). Following this qualification, ethnicity is mainly relevant to the degree that actors impose ethnic categories on others, which points to the central need to study the conditions under which ethnic categories become salient: why and under which conditions do people employ ethnic categories rather than others such as age, class or gender distinctions to conceptually organize persons and groups in their social environment?
There are different but complementary approaches to answering this question. One approach refers to micro-level social interactions. So far, many ethnographic studies, which are especially well suited for this task, have studied how ethnic categories are established, reproduced and negotiated in everyday interactions (for a review see Pachucki, Pendergrass and Lamont 2007). A second approach emphasizes the role of elites and sees ethnic categories 'as the outcome of a political and symbolic struggle over the categorical division of society' (Wimmer 2008, p. 985). Following this perspective, a growing literature engages in historical comparative investigations of the implementation of census categories (e.g. Kertzer and Arel 2002), or performs claim analyses of public media (e.g. Koopmans et al. 2005). A third approach points to the dependence of ethnic categorizations on wider societal and economic conditions. In this regard there exists a large body of survey research on the impact of economic decline or other societal conditions on anti-immigrant sentiments (for a review see Ceobanu and Escandell 2010). While survey research is well suited for such a comparative task, we cannot infer from the existing studies whether people use ethnic categories more frequently under any contextual conditions, because these studies rely on standardized items that presuppose certain categories. Analysing the use of ethnic categories should instead rely on open-ended questions, since only these may reveal whether ethnic categories are salient in the minds of respondents.
Given this background, the present work analyses an open-ended question placed in a recently conducted large-scale survey in Germany. In particular, this analysis deals with the categories native Germans apply to describe groups that are seen as being 'mostly responsible for problems in your neighbourhood'. In doing so, the inquiry makes two contributions. First, it offers novel insight on the ethnic folk classifications native Germans apply in organizing responsibility for problems in their neighbourhood. Germany is an interesting case for such an analysis, since Brubaker (1992) described it as a nearly ideal-typical case of a country with an ethnic conception of nationhood, which also characterizes its public discourse (Koopmans et al. 2005). Because ethnic categories are taken for granted in German public discourse, we should expect ethnic categories to be salient in everyday life and to be a manifest reference to the question who causes problems. This article shows, however, that members of the lower classes, drunkards, elderly and especially teenagers are more frequently blamed for neighbourhood problems than any ethnic minority.
Second, the inquiry aims to identify contextual factors that might explain why people characterize problem-groups in ethnic rather than other terms. Relying on implications of group threat theory (e.g. Blalock 1967) and the disintegration approach (e.g. Anhut and Heitmeyer 2000), I compare how out-group size and conditions of economic decline are associated with the likelihood that respondents use ethnic categories to characterize problem-groups in their neighbourhood. One key finding is that the effect of rising unemployment on the likelihood of seeing ethnic minorities as a problem-group increases with the initial level of unemployment and is thus accumulating in strength.

Theoretical background
Qualifying ethnicity as cognition does not answer what characterizes certain categories as ethnic. Following Weber (1987 [1921], p. 389), I define ethnicity as the subjective belief in a common descent, with the additional qualification that the denoted group must be larger than a family unit (Fearon 2003). Accordingly, ethnic categories can involve race as in the US, religion as in Northern Ireland, language as in Belgium, or nationality as in many immigrant-receiving Western European countries (Fearon 2003;Banton 2011). Given this definition, I wish to tackle two questions on the use of ethnic categories by native Germans when they describe neighbourhood problemgroups.

What kinds of ethnic categories do native Germans use?
Since ethnicity may involve such different things as language, race and religion, we may first wish to know what kinds of ethnic categories native Germans use to categorize ethnic out-groups. By discussing this question against the background of Germany's conception of nationhood (e.g. Brubaker 1992;Koopmans et al. 2005) and its immigration history (the following historical descriptions are based on Bade, Emmer and Lucassen 2010), I derive five different types of categories that are theoretically plausible.
According to Brubaker (1992), there exist important differences between states in the way they define their citizenries. Brubaker sees Germany as a nearly ideal-typical case of a state with an ethnic conception of nationhood, meaning that descent defines membership, rather than birth in the national territory. On this basis, Brubaker explains the way in which Germany dealt with post-World War II migrants, as reflected in its particularly low naturalization rates. Post-World War II migration to Germany started with guest workers, most of them coming from Turkey, Yugoslavia and Italy, who were recruited for the West German labour market from 1955 to 1973. These guest workers, their reunited families, and their offspring constituted about 8 per cent of the population and lived in Germany as foreign nationals Á whether born on German territory or not. In comparison, Aussiedler (immigrants from Eastern Europe, whose numbers rose by the mid-1980s) and, until reunification in 1990, Ü bersiedler (immigrants from the former GDR) had an unrestricted right to German citizenship until 1993, 1 because of their ethnic German descent. Koopmans et al. (2005) expand Brubaker's approach by showing that conceptions of nationhood do not only translate to the legal sphere, but also to countries' discursive opportunity structures, and thereby characterize their public discourse. Together, the tight association between nationhood and ethnicity and its manifestation in German public discourse suggest that for native Germans ethnic categories are characterized by nationality rather than by race, religion or language, and that ethnic out-groups are in general conceived of as foreign nationals. Hence, the first two types of categories I expect native Germans to use when describing ethnic out-groups are country of origin and foreign nationals categories.
In the mid-1980s the number of refugees from Eastern Europe, Turkey, Iraq and Yugoslavia who sought asylum in Germany increased rapidly. A peak was reached with over 400,000 applications for asylum in 1992. Accordingly, the German right-wing extremism of the early 1990s was characterized by a focus on asylum seekers (Braun and Koopmans 2010), who were seen as a main economic threat by the German population, even though they were not allowed to participate in the German labour market (Steinbach 2004). In consequence, the asylum law was reformed in 1993. Since then only those who do not enter over 'safe third countries' can apply for asylum. These events received a lot of media attention and established asylum seekers as a third type of category for ethnic out-groups.
Two recent trends contest this tradition of an ethnic conception of nationhood. First, political attempts to change German citizenship law began in the early 1990s. These attempts resulted in the citizenship law of 2000 which introduced the acquisition of German nationality by birth in the national territory as well as the right to naturalization after eight years of economically independent and legal residence. As a result, the country's foreigner politics changed to integration politics and the category of 'person with migration background' 2 was introduced into both the German Micro Census and the public discourse. Thereby, migrants are officially recognized as potential citizens rather than as tolerated foreign nationals, so that 'person with migration background' categories denote the fourth type of category native Germans might use when they try to be politically correct. A second recent trend in Germany's immigration history is the public attention Islam and Muslims receive. Since 9/11 there has been a public shift of attention from national groups to Muslims and Islam (Yurdakul 2009;Dolezal, Helbling and Hutter 2010). In contrast to the ethnic tradition of nationhood, this trend implies that ethnicity is increasingly defined by religion, suggesting religious categories as the fifth and final type.
However, against this elaboration of the conception of German nationhood as a background for this inquiry, Brubaker (1992, p. 242 n. 9) himself emphasizes that such conceptions are relevant primarily for national elites and questions whether they correspond to the views of the general population. Existing survey research on this question provides inconclusive evidence, with some finding relations between individuals' attitudes and citizenship models (e.g. Wright 2011), while others do not (e.g. Bail 2008). Such a line of critique seems even more valid when we disregard the particular ethnic categories used, and focus on the frequency with which we expect native Germans to use ethnic categories as compared to others such as class, age or gender categories. On the one hand, we should expect ethnic categories to be salient in everyday life and a manifest reference to the question who causes problems, given that ethnic categories are taken for granted in German public discourse and that media attention related to Islam and Muslims has been increasing in recent years (Dolezal, Helbling and Hutter 2010). Yet again, it is unclear how or whether one should expect these macro-level trends to translate to the frequency with which native Germans categorize groups that they see as responsible for neighbourhood problems. Accordingly, in Wimmer's (2004) ethnographic work on downtown neighbourhoods in Basel, Bern and Zurich, categories that denote ethnic out-groups are by no means the most frequently applied distinctions. I expect that while the kinds of categories people use reflect macro-level trends, the salience of categories is more context-specific.

Under which conditions do ethnic categories become salient?
But which are the context-specific conditions that make people characterize problem-groups in ethnic terms? While the literature on anti-immigrant sentiments has established a range of explanations (for reviews see Wimmer 1997; Ceobanu and Escandell 2010), I will here focus on two contextual factors and discuss these from the perspectives of, first, group threat theory and, second, the disintegration approach.
Group threat theory (e.g. Blalock 1967) assumes that larger percentages of minorities are associated with real or perceived competition for individual but also collective economic and symbolic resources as well as perceived conflicts over power. This, in turn, causes prejudices and growing social distances, because discrimination is a means in the competition and can ensure a dominant position in the struggle over power. For present purposes, this translates into the presumably obvious expectation that the frequency of categorizing problem-groups as ethnic increases with the percentage of ethnic minorities in the neighbourhood. But how should we imagine the relationship between the likelihood of using an ethnic distinction and ethnic minority group size to be shaped?
While the recent literature mostly tests a linear relation without much further discussion (for a review see Ceobanu and Escandell 2010), Blalock (1967) actually makes two different predictions. If discrimination is successful, the relation should be positive, but decreasing in strength. This follows from the idea that in areas with higher ethnic minority percentages discriminatory practices are already established, which compromises the minorities' competitiveness and thus makes further increases in the ethnic minority percentage less threatening (Blalock 1967, pp. 147Á50). For threats of power loss, Blalock assumes the opposite, namely a relation with an increasing slope, because he conceptualizes power as a multiplicative function of mobilization and resources (Blalock 1967, pp. 150Á4). Since both resources and mobilization potential increase with the percentage of ethnic minorities, minorities' power increases exponentially with their relative size and so should the majority's threat of power loss. Insofar as both processes overlap, we should expect what indeed most empirical studies on prejudice and threat test, namely an approximately linear relationship.
There is even good reason to assume a relationship that first increases and then decreases again. Two processes suggest such a relationship. First, Semyonov et al. (2004) emphasize the importance of perceived minority group sizes and how these are inflated, especially when there are only a few minority members. Second, inter-ethnic friendships and inter-marriages increase as opportunities to make contact rise with the percentage of ethnic minorities (e.g. Martinovic, van Tubergen and Maas 2009). Thus an initially positive relation between minority percentage and threat, aggravated by inflated perceptions of the latter, should level off and even start to decline as perceptions of out-group size become less inflated and processes of social integration emerge. Yet, Semyonov, Raijman and Gorodzeisky (2006) find no evidence for any curvilinear relation. Overall, what seems on first blush to be a fairly banal statement on the relationship between ethnic minority group size and the likelihood of categorizing problem-groups as ethnic is not so trivial, once we theorize about the actual shape of the relationship.
Next to the relative size of ethnic out-groups, group threat theory suggests that poor economic conditions foster ethnic conflicts, by making the resources over which ethnic groups compete scarcer. But as Blalock qualifies, only so far as 'discriminatory behaviour is perceived as instrumental either for large numbers of persons or for influential elites, in achieving status objectives by these efficient means ' (1967, p. 49). This qualification implies an interaction effect between poor economic conditions and the percentage of ethnic minorities, because in places with no or only a few out-group members, discrimination is not an effective means of competition within poorly performing economies; only a few people would be affected.
In comparison, the disintegration approach (e.g. Anhut and Heitmeyer 2000) does not make such a qualification in its prediction on the impact of economic decline on prejudices and discrimination. The approach builds on Durkheim's concept of anomie, which denotes a breakdown of a society's value system. It predicts that rapid social change will erode individuals' embeddedness in norms, values and social relations. This process of disintegration, meaning that individuals' capacities of adaptation cannot keep up with the speed of social change, is associated with substantive uncertainty, feelings of threat, and questions about personal, social and national identity. The responsibilities for these unsatisfying circumstances are then projected on ethnic out-groups that are seen as a threat, independently of any real existing ethnic conflict. Here discrimination is not a means to achieve status goals, but rather a coping strategy that helps to retain a positive self-conception, because the problem is transferred from one's own inability to adapt to the new circumstances to the 'outsiders', and also presents an opportunity to release frustration. Furthermore, nothing suggests a levelling of this process, in contrast to group threat theory's predictions on the effects of out-group size. Quite the contrary, a continuation of economic decline can function as validation or empirical support for actors who propagate ideas on the responsibility of ethnic minorities for the poor economic conditions. Thus a worsening of the economic conditions should yield more convincing power in regions with initially poor economic performance.
Analysing an open-ended question on neighbourhood problem-groups The EDCA Survey In order to pursue a quantitative analysis of the use of categories denoting ethnic out-groups, I analyse an open-ended question of the German subsample of the Ethnic Diversity and Collective Action Survey (EDCAS) (Schaeffer et al. 2011). The German part of the survey was conducted from October 2009 to April 2010 and consists of 7,500 standardized telephone interviews with participants who are at least 18 years of age. The survey has two oversamples of persons with a migration background. However, we should not intermingle the perspectives of the majority and migrant minorities, since these might differ and be driven by different factors (Wimmer 2004). Hence, the inquiry relies on the 4,581 interviews with native German respondents. The sample is stratified by fifty-five regions ('Kreise'), which are administrative units with an average population of about 190,000 inhabitants. In each region, sixty native German respondents were interviewed, and in five large cities (Berlin, Hamburg, Munich, Frankfurt and Duisburg) even 300 interviews were gathered. The comparison of these fifty-five regions allows for the investigation of context effects.

The open-ended variable and its coding
The initial third of the EDCAS entails a module on the neighbourhood. After having been asked about the general condition of the neighbourhood, but before being posed questions on immigrants and inter-ethnic relations in the neighbourhood, respondents were asked the following open-ended question: Which groups are mostly responsible for problems in your neighbourhood?You may now name up to three groups.
The interviewers were strictly instructed not to give any examples and to note the answers literally. The analysis draws on the various answers to this open question, which range from 'speeding car drivers ' and 'teenagers' to 'scouts'. In order to analyse this open-ended question, I had to code ethnic out-group distinctions made by respondents. Given the above definition of ethnicity, I accounted for national, racial, religious and linguistic dimensions. A frequent response was a negative ethnic category as in 'foreigners' or 'immigrants', which are negative definitions of not native German (negative in logical, not normative terms). In line with the theoretical discussion above, I differentiated three categories here: foreign nationals, persons with migration background and asylum seekers. The respondents also named concrete nationalities (such as 'Turks' or 'Albanians') with each receiving their own category in the coding procedure. Furthermore, there were instances in which respondents named racial categories (such as 'blacks'), regional categories (such as 'Southern Europeans') or religious categories (such as 'Muslims'), all of which again received their own category independent of their frequency. Finally, I coded the remaining answers, so as to give an impression of non-ethnic categories applied to characterize problem-groups. This latter coding, however, is not theoretically guided and just serves the purpose of a comparison with the ethnic categories. Some of these answers were not coded, because the categories they entailed were used by less than 0.5 per cent of the respondents who gave a valid answer; examples are 'scouts' or 'neighbours'.
It is important to note that many responses involved a couple of distinctions or a combination of categories, as in the example of 'Turkish teenagers' or 'young men with migration background'. In such cases the reply was coded as several categories and not as one category. This means the frequency of categories does not add up to 100 per cent, because people named up to three groups, and some of these responses involve several categories. This categorization process thus reveals how many respondents gave responses that involved the categories coded here.

Just another threat measure?
For the inquiry into contextual factors, I generated a binary variable differentiating those who used any ethnic category to characterize problem-groups in the neighbourhood from those who did not. To what extent is this binary variable not just another threat measure? Figure 1 shows the density distribution of a common threat measure and compares respondents who did draw an ethnic distinction with those who did not. The threat measure is a predicted factor score, which relies on four indicator variables, measuring economic, cultural and religious threat and whether migrant children threaten the quality of education native children receive. The dashed reference line shows the 75th percentile and the three-dot-dashed reference line the 25th percentile of the threat measure, here taken to designate the most and least threatened respondents. On average, respondents who did draw an ethnic distinction indeed feel more threatened (t-test (diff 50) p 00.0018). A larger share of respondents who did draw an ethnic distinction falls above the 75th percentile threshold (51 as compared to 39 per cent) and a smaller share below the 25th percentile threshold (12 as compared to 22 per cent).
But Figure 1 and the just-reported percentages also show that there is a considerable number of respondents who do not use an ethnic distinction yet feel very threatened; conversely, there is also a considerable number of people who do draw an ethnic distinction, but do not feel threatened. This supports my claim that the salience of ethnic distinctions in respondents' everyday lives measures something other than our standard threat measures, so that the current study complements existing research.

No answers, 'do not know'-answers and the population under investigation
Not all respondents did name a category. Some for example said there were no problems in their neighbourhood. In addition to respondents who gave such a valid response, a considerable number of respondents either did not give an answer or said they did not know an answer. Figure 2 shows that only about 31 per cent of the respondents did actually name a category, whereas about 17 per cent thought that no particular group could be blamed or that there were no problems in the first place. Overall, 46 per cent of the respondents gave a valid answer to the question, and their responses are investigated here. Fifty-four per cent of respondents not giving an answer or saying that they do not know an answer is, by normal standards, a high proportion of missing values. Yet, studying categories that people employ necessitates a vaguely formulated question, so as not to predetermine the answers. Accordingly, many people feel uneasy about misunderstanding the question and thus open-ended questions are known to generate higher rates of item non-response (e.g. Peterson 2000, p. 32).
Presumably, missing values are associated with selection bias. Let me make two comments on this point. First, mentioning a category could also be an indication of salience and a severe frustration with a problem-group. The threshold to actually blame a certain group in a public survey might be rather high. It is of interest, therefore, to study those who pass this threshold. Second, missing a value, i.e. not giving an answer or even explicitly mentioning not knowing one, might also indicate the absence of salient categories for responsible problemgroups. In this regard, additional analyses, which can be obtained from the author upon request, show that respondents who miss a value report levels of neighbourhood satisfaction that are much higher than and significantly different from those of respondents who did name a category. More importantly, those respondents missing a value are not appreciably more likely to move to another neighbourhood than respondents who said that there were no problems in their neighbourhood. This stands in contrast to respondents who did categorize a problem-group, since these are the only ones who do have a significantly higher likelihood of planning to move away. This supports my claim that respondents who miss a value can also be seen as an intermediate category between those saying that there are no problem-groups or problems and those saying that there are. For this reason, all inference statistics in the remainder of this study are conducted both without and with the 56 per cent of the respondents who did not give an answer, so as to further enhance the validity of the inquiry.

Estimation strategy
All inference statistics on contextual conditions are based on logistic regression models with clustered standard errors to account for clustering at the regional ('Kreis') level. In contrast to multi-level models, clustered standard errors assume 'no particular kind of withincluster correlation nor a particular form of heteroscedasticity' (Wooldridge 2003, p. 134). The models include several predictor variables: education, age, gender, religious denomination, regional population density, regional share of unemployed in 2007 and the change in the share of unemployed from 2007 to 2009, a dummy for East Germany, an interaction between the latter two variables and the share of foreign nationals. 3 To check the robustness of the findings, the regressions were estimated for three populations: for respondents who actually named any category and those who said there were no problem-groups or no problems in their neighbourhood (Model 1), only for respondents naming a category for a problem-group (Model 2), and finally for all respondents including those who did not reply or did not know an answer (Model 3). Since the results are by and large robust across all models, the discussion refers only to Model 1.

Results
To what extent do native Germans see ethnic out-groups as responsible for neighbourhood problems? Figure 3 shows the frequencies of ethnic and other categories mentioned by at least 0.5 per cent of the respondents who gave a valid answer. Taken together, about 13 per cent of those 46 per cent of respondents who did give a valid response used an ethnic category to denote problemgroups. But as in Wimmer's (2004) work, categories that denote ethnic  Turks out-groups are by no means the most frequent. About 23 per cent use teenagers as a category to describe problem-groups in the neighbourhood. One might be inclined to highlight that ethnic categorizations are the second most frequent when taken together. However, next to teenagers, responses concerned with the elderly, lower class, and drunkards are more frequent than the mentioning of Turks, the most frequent ethnic category. Animals, and among them mostly dogs, are more often seen as responsible for problems than the second most frequently mentioned ethnic category, namely Aussiedler. Even if we only take into account respondents who live in regions with shares of foreign nationals that lie above the median, the overall frequencies of ethnic categories increase (as can also be concluded from the regression models discussed below), but the discussed relational pattern does not change. These results suggest that people's everyday life issues differ from those that are expounded in public discourse. This is also true for other salient macro issues, such as class, religion and region. Neither divides between Catholics and Protestants or Christians and Muslims nor divides between people from East and West Germany play any role. Lower class is a more common category for problem-groups. Yet, examples of responses in this category include 'low educated', 'homeless people' or 'underclass mob' and these responses hardly reflect the conflict between capital and labour.
Turning from magnitude to type of ethnic distinctions, we first see that about half of the respondents who do use an ethnic category applied a distinction that denotes non-native Germans in general, such as 'migrant children' or 'foreigners'. A comparison of the three types of general distinctions (foreign nationals, persons with migration background and asylum seekers) shows that the foreign nationals category is more frequently applied than that of persons with migration background. This indicates that in their everyday life, large parts of the population did not follow the change in publicly used categories. The asylum seeker category was mentioned by only 0.36 per cent of the respondents, which implies that the conflict of the early 1990s has passed.
A slight majority of those respondents using an ethnic category specified a concrete ethnic group. Nearly thirty different categories describing distinct ethnic groups can be found in the answers, with only three being named by more than 0.5 per cent of the respondents. These three are Turks, followed by Aussiedler and Russians, and relate to the largest migrant groups in Germany. All other ethnic categories are mentioned by only a few, in all amounting to little more than 1 per cent of the respondents. Two-thirds of these are country of origin categories with the others being regional or national minority categories. A single respondent used the racial category blacks. This also means that less than 0.14 per cent of the respondents mentioned Muslims as a problem-group and thus used a religious category, and no one characterized an ethnic out-group by their language. One might argue that people answering 'Turks' are thinking of a religious distinction, since the categories of Turks and Muslims overlap to a large degree. But for this analysis it is precisely of interest which types of categories seem to be the more salient ones in describing certain people: country of origin, foreign nationals, persons with migration background or religious categories. The answer to this question is that the results strongly suggest the first three. As further validation we can investigate whether the usage of certain categories, such as Turks, is associated with other categories applied, such as Muslim. This is possible, because respondents could name up to three groups and because they could also name groups that involve several categories, such as 'Turkish teenagers'. Table 1 shows the tetrachoric correlation coefficients between the binary variables denoting the most frequently mentioned ethnic categories and those other mentioned categories, where the coefficients passed the 5 per cent significance threshold.
Respondents who denote problem-groups as foreign nationals also characterize problem-groups by their low socio-economic status, as in 'foreigners socially deprived' or as in the case of one respondent who named as a first group 'foreigners' and as a second 'welfare recipients'. The more politically correct respondents, i.e. those using the category of persons with migration background, have a tendency to characterize problem-groups as teenagers, as in 'teenagers with migration background'. If the 5 per cent significance level is considered, this also holds for those respondents who characterized ethnic out-groups as foreign nationals, but only to a weaker extent. Respondents who apply a category that denotes a concrete ethnic group are more likely to name a second concrete ethnic group. But neither is there a relation between the category of Muslim and any other category nor among the national categories that we tend to associate with Islam such as Turks and Moroccans. By and large, ethnic minorities are identified by their national origin and are not conceptually classified as religious, linguistic or even racial groups. Hence, ethnic categories largely correspond to the classical conception of German nationhood and reflect the recent political, legal and discursive changes only to a limited extent. After having discussed the types and frequency of ethnic categories, I now turn to investigate the conditions under which ethnic distinctions become salient: what makes native Germans blame ethnic minorities for problems in their neighbourhood? As any version of group threat theory would predict, the results of regression Model 1 of Table 2, which comprises the main results, show a highly significant positive relationship between the regional percentage of foreign nationals and the likelihood of mentioning an ethnic minority as being responsible for problems in the neighbourhood. As there are different expectations on the shape of this relation, it is visualized in Figure 4 along with its confidence intervals (denoted by the dashed lines). The figure shows the average marginal effect of the percentage of foreign nationals on the probability of seeing ethnic minorities as a problem-group conditional on the percentage of foreign nationals.
Neither Blalock's assumption of an exponential impact because of power threat (as a linear slope would signify) nor the expectation of a linear relationship (as a flat line above the zero level would suggest) are supported. Instead, we see the following curvilinear relationship: initially, as the percentage of foreign nationals increases, the marginal effect of the percentage of foreign nationals on the probability of drawing an ethnic distinction increases as well. After a maximum turning point at about 11 per cent foreign nationals (first grey line from the left), the positive marginal effect decreases in strength. At about 23 per cent foreign nationals (second grey line from the left), the model suggests an inflection point, where the relationship even turns from positive to negative. However, here the observations get sparse and thus the confidence intervals are too broad to allow for any thorough conclusions. Overall, these findings fit best with the fourth assumption. According to this interpretation, inflated perceptions of out-group size are driving the increasing marginal effect, as there are few ethnic minority members initially. After a certain number of ethnic minority members have established themselves in the region, the perceptions become more realistic so that the marginal effect remains positive, but decreases. Eventually, processes of social integration diminish this marginal effect even further.
Turning to economic decline, group threat theory conditions any association with the share of ethnic minorities, since only in their presence is discrimination an efficient means of achieving status goals. However, Model 1 of Table 3, which shows the results of the respective additional regressions, does not support the hypothesis on an interaction effect between an increase in unemployment and the percentage of foreign nationals. Maybe, as the disintegration approach suggests, the assumption of discrimination as an instrumental means is too rigid?  Focusing on Model 1 of Table 2 again, we see the effect of rising regional unemployment conditioned on the initial level of unemployment. The single coefficients should not be interpreted, because of the interaction term, which makes the impact of rising unemployment dependent on the initial level of unemployment. Instead, Figure 5 visualizes the results, by showing the average marginal effect of the change in unemployment as a function of the initial level of unemployment.
The figure suggests that for regions with initially low levels of unemployment an increase in unemployment has no impact on the probability of categorizing problem-groups in the neighbourhood as ethnic. But as we turn to regions with initially higher levels of unemployment, the impact of an increase in unemployment becomes stronger. In regions with an initial level of unemployment of more than 10 per cent, an increase in unemployment is significantly related to an increase in the probability of seeing ethnic minorities as responsible for neighbourhood problems. With higher levels of initial unemployment the marginal effect of an increase in unemployment accumulates in strength. Thus the marginal effect of economic decline behaves in the opposite way to that of out-group size: while the marginal effect of out-group size levels off and may even become negative, the impact of economic decline seems to accumulate in strength.

Conclusion
As a consequence of the claim that ethnic categories are first and foremost cognitive distinctions that actors impose on the world, the central need arises to study the conditions under which these categories become salient. Given this background, this study sought to investigate two questions: first, along which ethnic distinctions, do native Germans conceptually organize responsibility for problems in their neighbourhood, and, second, how do contextual conditions help explain why people characterize problem-groups as ethnic over other categorizations? To do so, I analysed answers to an open-ended question asking respondents about 'which groups are mostly responsible for problems in your neighbourhood'. Such an open-ended item is not just another threat measure, because it allows investigating the salience of ethnic categories in the minds of respondents, something existing survey research, which relies on standardized survey items, does not inform us about.
Regarding native Germans' use of ethnic folk classifications, this study shows that about 13 per cent of the respondents did draw an ethnic distinction to denote neighbourhood problem-groups. Yet the elderly, drunkards, lower-class members and especially teenagers are mentioned more frequently than any single ethnic category. Thereby, this study suggests that the issues of people's everyday lives differ from those that are expounded in public discourse. About half of the respondents who used an ethnic category applied a general category by drawing a distinction against non-natives in general. The other respondents who did draw an ethnic distinction used country of origin categories. In contrast to Brubaker's doubts, this study suggests that the ethnic distinctions native Germans draw do reflect the classical conception of German nationhood and its immigration history. In so far as Banton (2011) is right in positing that the way the majority conceives of itself provides the standard against which members of minorities are judged, this also allows us to infer that the selfconception of native Germans is national and not religious or racial. To further validate this conclusion, it would be interesting to replicate this study in countries with different conceptions of nationhood and immigration histories, such as France, the UK, the US or the Netherlands, to see whether in these countries ethnic out-groups are defined in other terms, such as religion or race.
The recent legal, political and discursive changes are only partially reflected in the everyday ethnic category usage of native Germans. Only half of those respondents who use a general category follow the change in official categories, which recognizes ethnic minorities as migrants and hence as potential citizens rather than tolerated foreign nationals. Furthermore, despite growing media attention, Islam and Muslims play no role in the ways native Germans ascribe responsibility for neighbourhood problems. Further research is needed to investigate whether certain kinds of ethnic distinctions translate more easily from the macro level to everyday life, and how they gain in salience. In this regard, it would also be interesting to study migrants' use of ethnic categories and investigate whether it relates to the public discourse of the host country or the country of origin.
With respect to the two explanatory contextual determinants, the results suggest a curvilinear relationship between out-group size and seeing ethnic minorities as problem-groups, which first increases in strength, but then decreases in strength after a threshold of about 11 per cent. After this threshold, processes of social integration and less inflated perceptions of out-group sizes presumably occur; this would explain the specific curvilinear function. In contrast, the impact of regional economic decline, as measured by change in regional unemployment, increases in strength in regions with initially higher levels of unemployment. In such regions, according to one possible explanation, the negative development can be framed as empirical validation of the claim it was the minorities' fault. Overall this results in the sobering conclusion that while people do adapt to ethnic diversity, the negative impact of economic decline is accumulating in strength.
These findings complement existing research in three important ways. First, they show that predictions of central theories on prejudice also hold for the salience of ethnic distinctions. Second, rather than just testing for linear dependence, the study shows more complex and theoretically plausible relations between ethnic out-group size, economic decline and the probability of blaming ethnic out-groups for neighbourhood problems. Third and most importantly, however, the study answers the question raised by the literature on ethnicity as cognition, why and under which conditions people start to employ ethnic categories rather than others such as age, class or gender distinctions to conceptually organize persons and groups in their social environment; something ethnographic studies, claim analysis and other research have attended to, but that existing survey research has not been able to do.
The study also faces limitations. Challenges of analysing nonstandardized variables for category usage include dealing with larger numbers of missing values and sensitivity of the results to the coding procedure. Several measures, such as regression models with different populations or comparisons to other indicators, were taken to counter possible objections. Despite these problems, the fact that such an open-ended variable enables insights into people's lines of reasoning without conceptually predetermining categories seems to make for a rewarding social science approach. With powerful software to handle string variables, future research should investigate the potentials of open-ended questions and different ways to code such data in both conceptually meaningful and empirically reliable ways.