Painting All Foreigners With One Brush? How the Salience of Muslims and Refugees Shapes Judgements

Attitudes towards foreigners are widely researched, most frequently in survey studies. However, in that context it is often unclear which attitude object respondents have in mind, and thus what their answers refer to. This paper uses a representative sample of 3,195 Germans who reported which groups they think of when thinking about foreigners living in Germany. We found that Germans disproportionately think of groups who are Muslim, and that such salience is associated with more negative attitudes towards “foreigners.” This holds true when controlling for attitudes towards Muslims; in fact, thinking of Muslim groups when thinking about foreigners moderates the relationship between anti-Muslim and anti-foreigner attitudes. The relationships were weaker when respondents think of Turks, a large and long-standing minority in Germany, suggesting an attenuation of the links through familiarity or intergroup contacts. No relationship was found between thinking of refugees and attitudes towards foreigners. Implications for research are discussed, particularly regarding the interpretation of self-reported attitudes towards foreigners and the study of populist strategies.


Non-Technical Summary Background
People's attitudes towards foreigners matter; they shape policy and public debates, and arguable also fuel conflict between social groups. Such attitudes are often measured by simply asking people how they feel about 'foreigners'. However, it is typically unclear who the respondents think of in making their assessments. If they only think of specific groups, and if these groups differ systematically from the foreign population in a country, then the reported attitudes might be shaped by the presence of specific groups in the minds of the respondents (their "mental presence") as much as by attitudes towards any one group. In addition, in such case, attitudes towards foreigners could be manipulated by changing the mental presence of some groups.

Why was this study done?
Earlier research in different countries had suggested that in surveys, the foreigners that come to mind first for respondents are associated with their attitudes towards all foreigners. We now wanted to replicate that research and test three things: whether these findings also holds when all groups that respondents had in mind are considered rather than just the first, if thinking of multiple groups (and thereby having a more diverse mental representation of foreigners) has a positive effect and whether the effect appears because attitudes towards the specific groups that are mentally present are then generalized towards all foreigners. We focused on the effect of thinking of refugees and of Muslims as those were the two most relevant dimension of cultural distance in Germany at the time when the research was conducted.

What did the researchers do and find?
We based our research on data from the German General Social Science Survey (ALLBUS) that surveyed a representative sample of 3,195 German citizens in 2016 about their attitudes towards foreigners and asked what groups they think of when thinking of 'foreigners. ' We found that Muslims are much more present in respondents' minds than in Germany's foreign population, in fact, 81% of respondents first thought of a Muslim group while Muslims make up 37% of Germany's foreign population. We then analyzed whether people who thought of Muslim groups or of refugees reported more negative attitudes than those who thought of less culturally distant foreigners. Those who thought of Muslims indeed reported more negative attitudes towards foreigners overall, particularly when they thought of groups other than Turkish people. Contrary to our expectations, thinking of refugees was not generally associated with attitudes towards foreigners, only thinking specifically of 'asylum seekers' was associated with more negative reported attitudes towards all foreigners. Thinking of a greater number of groups was consistently associated with more positive attitudes. Finally, given that anti-Muslim attitudes are stronger and more widespread than anti-foreigner attitudes, we tested whether the negative effect of thinking of Muslims appears because anti-Muslim attitudes are generalized to all foreigners when respondents think of Muslims as typical foreigners. This indeed appeared to be the case.
Research on xenophobia, prejudice and policy preferences frequently asks respondents to report their attitude towards foreigners, without further defining or specifying the target group, based on a tacit assumption that people can and do refer to a uniform and stable attitude object (see Asbrock, Lemmer, Becker, Koller, & Wagner, 2014, for a list of such studies). However, it is unclear to what extent such an attitude object exists. Indeed, it has been shown that when responding to questions about immigrants, only between one fifth and one third of respondents report that they thought of immigrants in general; most others thought of a specific salient group, often the largest group in the country under consideration (Braun, Behr, & Kaczmirek, 2013).
From previous research, we know that attitudes towards foreigners (Asbrock et al., 2014), immigrants (Blinder, 2015) and strangers (Spruyt, van der Noll, & Vandenbossche, 2016) are influenced by the specific groups that first spring to mind for the respondent. This paper adds to this research by confirming the effect with the coding of an unrestricted number of open-ended answers, thereby adding insights into the effect of heterogeneity among the salient groups. In addition, we show that the associations hold when controlling for the specific attitudes towards Muslims, and that familiarity with a distant group appears to attenuate the effect. Lastly, we highlight that the attitude objects that our representative sample of respondents referred to diverge from the composition of the German foreign population, which indicates that self-reported attitudes towards foreigners are likely not targeted at the actual foreign population. To situate this work, a brief review of the process of forming and expressing attitudes towards complex categories is in order, before the specific literature in this field is considered.

Judgements About Complex Categories
When asked to express an attitude towards a complex category, respondents need to develop a mental representation of the attitude object (Blinder, 2015). This is often guided by the availability heuristic, in which case the nature of a category is deduced from the instances that most readily spring to mind (Tversky & Kahneman, 1973). Alternatively, a complex category can come to be represented by an exemplar that is seen as prototypical (Smith & Zárate, 1992). In that case, judgements about the category flow directly from judgements about exemplars. This has been shown to play a large part in the formation of judgements about social categories, both in terms of their valence and their stability (Sia, Lord, Blessum, Ratcliff, & Lepper, 1997). In a sense, these processes are obvious and complementary -as Lakoff (1975) reported, most people are intuitively both willing and able to assign different degrees of 'birdiness' to different birds, deeming robins to be very birdy, ducks to be somewhat birdy and penguins not to be very birdy at all. Consequently, it can be expected that most people primarily reference very birdy birds that spring to mind readily when thinking about the category, so that their judgements might differ from those they would reach when systematically considering all category members (cf. Blinder, 2015).
In the context of foreigners, the English language explicitly suggests the same opportunity for a gradation of foreignness. The dictionary sense of foreign ("strange and unfamiliar", according to Oxforddictionaries.com, 2019) allows us to consider some foreigner as more foreign than others. If such culturally distant groups are more likely to spring to mind when thinking about foreigners, this is likely to shape judgements about the overall category, because attitudes tend to become more negative as one considers outgroups with greater perceived cultural distance from the ingroup (Steinbach, 2004).

The Influence of Outgroup Heterogeneity
Foreigners, as a category, can come to be represented in the respondents' mind by one or several salient groups, i.e. groups that have a higher "probability of invocation" when considering a superordinate set (Stryker & Serpe, 1994). A greater number of salient groups would indicate a greater extent of perceived outgroup heterogeneity, which has been widely suggested to be associated with more positive attitudes towards the outgroup (Brauer & Er-rafiy, 2011). This is, for instance, because a complex representation of a category is associated with less intense affect towards the category (Linville, 1985), and because people are slower to make judgements about heterogeneous rather than homogeneous groups (Lambert, 1995). More recent experimental work has indeed shown that an increase in outgroup heterogeneity leads to less prejudiced attitudes and less discriminatory behaviors (Brauer & Er-rafiy, 2011;Er-rafiy & Brauer, 2013), so that the heterogeneity of the representation of foreigners by respondents can be expected to influence attitudes towards foreigners alongside the nature of the salient groups.

Salience of Groups and Attitudes Towards Foreigners
Research in several countries has shown that attitudes towards foreigners are indeed shaped by the specific groups that respondents have in mind when thinking about foreigners. Of course, this higher probability of evocation, which we will term group salience, draws its relevance from an interaction with category salience, given that any one specific group (e.g., Somalis) can be subsumed into multiple categories (e.g., Muslims, refugees, Blacks), which might trigger different attitudes; nevertheless, the question which groups are salient in the first place deserves consideration. This has been researched in multiple contexts to date. Considering British public opinion, Blinder (2015) showed that thinking of long-term immigrants from EU countries was associated with particularly negative attitudes towards immigration overall. In a study in Flanders, people who primarily thought of Muslims or people from predominantly Muslim countries reported the highest level of prejudice against strangers (Spruyt, van der Noll, & Vandenbossche, 2016). Finally, in the German context, Asbrock and colleagues (2014) found that thinking of Turkish people first was associated with more negative attitudes towards foreigners than thinking of other groups of immigrants. However, that effect mostly emerged in the East of Germany, where Turkish foreigners are comparatively rare, suggesting that salience in the mind of the respondents may be particularly relevant in as far as it diverges from the composition of the foreign population that is in fact present. This might be because in such cases contact with foreigners is likely not to occur with members of the most salient groups, which reduces its potential to improve generalized outgroup attitudes (Pettigrew & Tropp, 2011).
Overall, the research to date suggests that the salience of foreigner groups needs to be considered alongside any assessment of attitudes towards the entire category. However, most research to date either asked respondents to choose among given categories (e.g., Blinder, 2015), which is unsuitable to capture all salient groups, or asked open questions, but then considered only the first mention rather than all responses (Asbrock et al., 2014;Spruyt & van der Noll, 2017;Spruyt et al., 2016). The first mention is likely to represent the most easily accessible group, which can be expected to have the greatest impact on judgements (Schwarz, 1998); however, by considering all responses we will be able to test that assumption and consider the impact of heterogeneity within the responses, which will strengthen the knowledge base. Additionally, none of the studies to date have considered the specific attitudes towards the most salient groups, thereby omitting to test the mechanism that we expect to underlie the salience effect: the generalization of attitudes from the salient group(s) to the overall category. We address this by testing whether salience moderates the link between group-specific attitudes and attitudes towards all foreigners.
With regard to such a generalization of attitudes, it was suggested above that attitudes towards the category are derived from judgements about the salient exemplars; additionally, much research in the intergroup contact context has shown that attitude changes towards one group are generalized towards other groups that are perceived as similar (Harwood, Paolini, Joyce, Rubin, & Arroyo, 2011;Pettigrew, 2009), which we would expect to apply to a group ("foreigners") and its most salient sub-groups. Therefore, seeing foreigners as Muslim, for instance, should moderate the association between anti-Muslim attitudes and anti-foreigner prejudice. Given that anti-Muslim attitudes tend to be among the strongest categories of prejudice (Heath & Richards, 2019;Statham, 2016;Strabac & Listhaug, 2008), this could explain why thinking of Muslims would be associated with particularly negative attitudes towards foreigners. Here again, the present study adds to the literature by testing this pathway explicitly. Furthermore, we use the specific German context to explore to what extent familiarity with a culturally distant group, in this case Turkish immigrants, attenuates the effect of cultural distance on generalized attitudes towards foreigners.

The German Context
Historically, immigration flows differed strongly between West and East Germany. During the 1960s and 1970s, West Germany strove to attract short term workers, the so-called Gastarbeiter, who were primarily recruited in Turkey and Italy (Meier-Braun, 2015). Max Frisch pithily summarized the experience: "We asked for workers, but human beings came;" many brought their families and settled (Frisch, 1965, as cited in Meier-Braun, 2015).
The East, on the other hand, exclusively recruited workers from other socialist countries, and enforced that they left the country after five years, so that by the time of reunification in 1990, only 200,000 foreigners lived in the entire German Democratic Republic (Bade & Oltmer, 2004). This history is still visible in the current foreign population; by the end of 2017, 4.4% of the population in East Germany held a foreign citizenship, while this was the case for 12.6% in West Germany; in the West, Turkey, Poland and Italy were the most common countries of origin, while Poland, Russia and Syria occupied the top positions in the East (Destatis, 2017). In line with the lower share of foreigners, attitudes towards foreigners tend to be more negative in East Germany (Asbrock et al., 2014), partly because intergroup contact is less frequent (Wagner, van Dick, Pettigrew, & Christ, 2003).
Within Germany, different groups of immigrants are perceived in diverging ways. Recently, Froehlich and Schulte (2019) asked Germans to rate 17 immigrant groups on warmth and competence and found that they fell into four distinct clusters; the largest consisted of immigrants from comparatively poor and mostly majority-Muslim countries that were stereotyped as low in competence and moderate in warmth. Accordingly, we aimed to test the impact of thinking of Muslim groups on the attitudes towards all foreigners.
We additionally considered the impact of thinking of refugees, particularly because the data presented in this paper was collected during the summer of 2016, towards the end of the largest arrival of refugees in German history.
In 2015, 890,000 asylum seekers had arrived in Germany, mostly from Syria, while another 280,000 arrived in 2016, which together corresponded to 12.2% of the entire foreign population in Germany (BAMF, 2017). Recent work has shown that the way German citizens perceive them can strongly depend on the nuances of language -in a survey experiment, 'refugees', for instance, were perceived much more positively than 'asylum seekers', Journal of Social and Political Psychology 2020, Vol. 8(1), 246-265 https://doi.org/10.5964/jspp.v8i1.1283 Painting All Foreigners With One Brush 250 partly because different motivations were ascribed to them (Kotzur, Forsbach, & Wagner, 2017). In our data, both Muslims and refugees were modally seen to refer to Syrians, so that reversely, any impact of thinking of Syrians as a salient group of foreigners might be moderated by the specific category that becomes salient. This might limit our ability to find an effect of the group salience of refugees alone.

The Present Research
As has been demonstrated, it is important to understand who respondents think of when they think of foreigners and how these spontaneous associations affect evaluation of foreigners. Therefore, we explored how the foreigners Germans have in mind, i.e. the most salient groups, compare to the actual foreign population in Germany, and then tested in how far this is associated with attitudes towards foreigners overall. In the coding of the open-ended responses, we focused on the refugee-status and the majority religion of the groups under consideration, because these appear to have been two dominant dimensions of distance in the given context and because measures of respondent's attitudes towards these groups were available in the dataset. After initial descriptive analyses, we tested four hypotheses:

H1:
Having greater heterogeneity within one's mental representation of foreigners (i.e. mentioning more groups) would be associated with more positive attitudes towards foreigners.

H2:
Thinking of Muslim groups and of refugees when thinking of foreigners would be associated with more negative attitudes towards foreigners overall.

H3:
The effect of thinking of Muslims and refugees would be particularly strong when these are the only salient groups. Where other groups are mentioned, the effect would be stronger when Muslims or refugees are mentioned first than when they are mentioned further down the list.

H4:
The effects of thinking of Muslims would be weaker when respondents think of Turkish people. This would particularly be the case in West Germany.
Hypothesis 4 was tested to reflect the fact that a Muslim group, namely Turkish people, constitutes the largest and longest-established group of foreigners in Germany, particularly in West Germany where contact has been occurring for half a century.

H5:
Thinking of Muslims and refugees would moderate (strengthen) the association between attitudes towards Muslims and refuges and attitudes towards foreigners generally.

Method Dataset
This paper is based on data from the German General Social Survey (ALLBUS) 2016 (GESIS, 2017) that focused on attitudes towards immigrants and minorities in Germany. ALLBUS employs a random cluster sampling approach of residents of Germany above the age of 18 that covered 162 sample points and a total of 3,490 respondents in 2016, with a purposive oversampling of respondents from East Germany. The data was collected through computer assisted personal interviewing (CAPI) during the summer of 2016, which achieved a response rate of 34.6%.
For all analyses in this paper, it was weighted in line with the guidance in the variable report. In order to capture

Variables Independent Variable: Salient Foreigner Groups
ALLBUS 2016 included an open question that ran "which groups do you think of when you think of foreigners living in Germany?" This was coded by the survey team using a detailed scheme (Wasmer, Kwasniok, & Kialunda, 2018); for this paper, we summarized the codes on two dimensions: the majority religion of the country or region in question, and whether the code denoted foreigners who came as refugees.
The classification of religion, based on data from the Pew Research Center (2012), was not possible for answer codes that either referred to general groups (e.g., "labor migrants") or to regions with no clear majority religion (e.g., "Africa"); in the 10.1% of cases where respondents exclusively referred to such groups, this variable was coded as missing. Responses were coded as referring to refugees when they either explicitly mentioned refugees or asylum seekers or when they referred to countries for which over two thirds of nationals resident in Germany are officially recognized as refugees (i.e. Syria, Afghanistan, Iraq, Somalia and Eritrea; based on Destatis, 2017).
Respondents were not limited in the number of groups they mentioned; 35% of respondents mentioned only one group, while 12 respondents mentioned more than ten groups. We coded whether all or none of the groups referred to Muslims or refugees, respectively, and in the case of mixed responses, whether Muslims or refugees were mentioned first or later in the list. This yielded four categories. In the case of religion, we additionally separated mentions of Turkish people from those of other Muslims. The resulting categories and the distribution of respondents across categories are shown in Table 1 below.

Dependent Variable: Attitude Towards Foreigners
This was measured with ten items that asked respondents to what extent they agreed, namely "foreigners … burden our welfare system, take away jobs, commit more crimes, depress the academic level in schools, cause a loss in social cohesion through their presence, support pensions, enrich our culture, create jobs, make Germany more tolerant and open through their presence, reduce the skill shortage," with the last five items being reversecoded. It used a seven-point Likert-scale (1 = do not agree at all, 7 = agree entirely), with Cronbach's α = .86.

Supplementary Variables
Attitudes towards refugees and Muslims -To further understand how group salience affects generalized attitudes, we tested whether the salience of specific groups moderates the link between attitudes towards these groups and foreigners overall. Regarding attitudes towards refugees, respondents were asked to think about social changes in Germany in the coming years and to rate whether refugees presented more of an opportunity or more of a risk to the welfare state, public safety, social cohesion and the economy. These used a five-point scale (

Painting All Foreigners With One Brush 252
Outgroup heterogeneity -The number of groups mentioned in response to the question about the foreigners that come to mind was included as an indicator of outgroup heterogeneity. It ranged from 1 to 17, with 94% of responses falling between 1 and 5. To prevent the potential for outliers to distort results, it was capped at 8, three standard deviations above the mean. This affected 32 responses.
Demographic covariates -Respondents' age in years, gender and level of educational attainment were included as covariates. Gender was coded as a binary variable by the interviewer without the respondents' involvement, while education was treated as a continuous variable based on the ISCED 1997 levels, from 1 = basic education to 6 = second stage of tertiary education. Respondents who were still attending school at the age of 18 (n = 22) were coded on level 3 as having completed upper secondary education.

Missing Data and Effective Sample Size
Respondents who had missing data on any of the variables under consideration were excluded from the dataset.
This was most commonly the case for the open question, where 291 respondents (9.0%) gave no valid response, mostly because they did not refer to any identifiable group. A dummy variable indicating this missingness was not correlated with the attitudes towards foreigners, refugees or Muslims, ps > .36, so that no corrections were applied.
After this exclusion, all remaining respondents provided valid answers for both dependent variables. Regarding the covariates, 16 (0.5%) did not fully report their attitudes towards Muslims and 13 (0.4%) did not complete all items regarding their attitudes towards refugees; they were excluded. Finally, four participants were excluded because they did not report their age and 1 participant because they did not report their educational attainment.
These exclusions yielded a sample size of 2,875 for our analyses with regard to refugees and 2,586 for those with regard to religion, given that 289 participants (10.0%) mentioned no groups that could be associated with a majority religion.

Coding Framework, Replication Code and Preregistration
Our full analysis code in R as well as the coding framework can be found on the Open Science Foundation repository: https://osf.io/tzjxp/. An earlier version of the study was pre-registered on AsPredicted.org; however, the review process resulted in major changes with regard to the analytic approach. i

Descriptive Analyses: The Groups That Come to Mind First
Unsurprisingly, Turkish people as the largest and one of the longest-established national group of foreigners living in Germany were the most salient group in the sample. Overall, 52.5% of respondents mentioned them first, including 44.5% in East Germany and 54.3% in West Germany, while an additional 15.0% mentioned them further down their list of responses. However, Turkish people only make up 14.9% of foreigners living in Germany, while most foreigners actually hail from majority-Christian countries (Destatis, 2017). Figure 1 shows the distribution of the salient groups and the make-up of Germany's foreign population by religion and refugee status.
For refugees, the overall share of mentions was roughly in line with their share in the populations. However, out of those respondents who did not mention Turkish people first, 44.8% reported a refugee group as the most salient group, while refugees made up 17.6% of the foreign population in Germany. Overall, the data showed that Germans mostly think of Muslim groups when thinking of foreigners, in marked contrast to the make-up of Germany's foreign population. Refugees were also over-represented among the salient groups, yet to a lesser extent than one might have expected in light of the strong media coverage of the 'refugee crisis' during the period of data collection. While it was not surprising that salience is driven by factors beyond mere frequency, these initial analyses indicate that for Germans, the foreigners that come to mind differ rather strongly from the foreigners that are in one's country.

Association of the Groups Mentioned With Attitudes to Foreigners
Preliminary analyses were conducted to consider the correlations between the variables, these can be found alongside descriptive statistics in Table 2. All attitude measures are coded so that higher numbers indicate more negative attitudes. The correlations indicated that, in line with expectations, outgroup heterogeneity was associated with more positive attitudes across all measures (Hypothesis 1). Additionally, all attitude measures were highly correlated. Among the demographic covariates, age predicted greater prejudice, while being from West Germany and being more educated predicted lower prejudice. Gender was only significantly related to attitudes towards Muslims where females reported slightly more negative attitudes. Regression analyses were carried out to test whether the religion and refugee status of the groups mentioned was associated with attitudes towards foreigners (Hypothesis 2), whether the order or groups mattered (Hypothesis 3) and whether thinking of Turkish people had a distinct effect compared to thinking of other Muslims (Hypothesis 4). For that, the categories of mentions were dummy-coded; the categories and the distribution of mentions across them can be found in Table 1. In order to control for the effect of perceived outgroup heterogeneity, the number of groups mentioned was included as a separate predictor in the models, alongside the demographic covariates.

Religion of Salient Groups
A regression model was run to predict attitudes towards foreigners from the religious composition of the salient groups, controlling for outgroup heterogeneity and the demographic covariates. The overall model was highly significant, F(11, 2589) = 29.93, p < .001, R 2 = .11; coefficients are shown in Table 3. Mentioning one or multiple Muslim groups was associated with greater anti-foreigner prejudice. This was strongest when Muslim groups made up the entire list or were at its start, and when they were not Turkish. When Muslim groups were listed further down a diverse list, or when only Turkish Muslims were among the salient groups, the effect was smaller. Pairwise comparisons between the conditions are shown with superscript letters in Table 3; they were obtained by rerunning the regression model with alternative reference categories, recording the dummy variables' significance and then creating the letter representation (Piepho, 2004
To explore how this effect obtains, we next tested a moderation model to see whether the salience of Muslim groups strengthens the link between attitudes towards Muslims and attitudes towards all foreigners (Hypothesis 5). When adding attitudes towards Muslims to the regression model, the interaction term between the category of mentions and the attitudes towards Muslims was significant, F(6, 2582) = 2.60, p = .016; the resulting regression models is shown in Table 4. The results suggest that thinking of Muslims moderates and strengthens the link between attitudes towards Muslims and attitudes towards foreigners across the categories, though the interaction is only significant for those who think of Muslims (incl. Turkish Muslims) only. Figure 2 shows that the gap is particularly prevalent at high levels of negative attitudes towards Muslims -here, thinking of them as the most salient foreigners appears to strengthen the generalization of anti-Muslim attitudes into anti-foreigner attitudes.
As a supplementary, exploratory analysis, we tested whether more negative attitudes towards Muslims were associated with greater salience of Muslim groups when thinking of foreigners. To that effect, we ran a multinomial logistic regression to see whether respondents' attitude towards Muslims would change their likelihood of mentioning Muslims among their salient groups, controlling for their region, age, gender and education. The resulting odds ratios are shown in Table 5 and indicate that more negative attitudes towards Muslims were indeed associated with a greater salience of Muslim groups, as was being younger; gender and education had no effect. In line with the population ratios, Turkish people were more salient in West Germany. Even though the salience of Muslims was associated with attitudes towards Muslims, the composition of salient groups explained a significant additional share of variance; the regression model in Table 3 explained significantly more variance than the same model with the salient groups removed, ΔR 2 =.008, F(12, 2582) = 3.49, p < .001.

Refugee Status
Contrary to our expectations, whether respondents thought of refugees was not associated with their attitudes towards foreigners. The model including that variable did not differ from the base model including only the number of groups mentioned and the region, age, gender and education of the respondent, ΔR 2 = .002, F(3, 2880) = 1.86, p = .13. To understand whether this null finding could be explained by the different associations evoked by different terms, we ran an exploratory analysis in which we created dummy variables that indicated whether respondents had mentioned one of the three most commonly cited groups of refugees (Syrians, Iraqis and Afghans) or generally referred to 'refugees' or 'asylum seekers' (neglecting the order of mentions). The resulting regression model was highly significant, F(8, 2880) = 41.11, p < .001, and is shown in Table 6. It supports the finding by Kotzur and colleagues (2017) that 'asylum seekers' are thought of particularly negatively. It also showed that the salience of Muslim refugees was associated with somewhat worse attitudes towards foreigners (likely primarily because they are Muslim), while just thinking of 'refugees' was not associated with attitudes towards foreigners.

Discussion
Our results generally lend support to the hypotheses, with a few notable deviations from expectations. Initially, the descriptive analyses showed that Germans tended to think of disproportionately foreign groups when they think of foreigners living in Germany: Muslims and refugees featured more prominently in the respondents' minds than in Germany's foreign population. Therefore, it was important to consider whether this salience affects selfreported attitudes towards "foreigners" as that would imply that they do not match attitudes held towards the actual foreign population. This is indeed the case when respondents think of Muslim groups as the most salient foreigners; when non-Turkish Muslims were mentioned, exclusively or as the first group among a diverse set of responses, the respondents reported much more negative attitudes towards foreigners. The effect was weaker when Muslims featured further down the list of responses. It is also weaker when respondents only thought of Turkish Muslims, likely because there have been opportunities for intergroup contact for decades (e.g., Wagner, Hewstone, & Machleit, 1989). For refugees, however, there were no such associations; against our expectations, thinking of refugees as salient groups of foreigners does not predict attitudes towards foreigners overall. This is likely because for refugee groups, category salience is of particular importance. While we cannot systematically test this with our data, we have replicated and extended the finding by Kotzur, Forsbach, and Wagner (2017) by showing that while respondents who thought of 'asylum seekers' reported much more negative attitudes towards foreigners, thinking of 'refugees' was not associated with overall attitudes towards foreigners, When it comes to explaining the association, we show that thinking of Muslims moderates and strengthens the link between attitudes towards Muslims and attitudes towards foreigners. This generalization effect has important implications for the study of populist strategies, which will be considered below.

Stability of the Attitude Objects
Considering the finding that self-reported attitudes towards foreigners depend on the specific attitude objects that come to mind, it is important to consider how stable these objects are. While our study was based on observational and cross-sectional data, it can yield two indications. Firstly, when comparing our results to those reported by Asbrock and colleagues (2014), it appears that the salience of Turkish people has decreased over time. Their data, collected between 2002 and 2004, showed that 58.3% of respondents thought of Turkish people first, while this was the case for 53.0% in the ALLBUS 2016 data reported here. This accords with increasing diversity in the foreign population in Germany, but also suggests that the attitude objects referred to by survey respondents are becoming more differentiated, thus increasing the challenge to interpretation.
Secondly, mentions of refugees were somewhat less frequent than expected at 22.0% of all mentions, even though refugees made up a greater share of the German foreign population than Turkish people in 2016 (17.6%) and their influx was highly present in the media. However, in the data presented by Asbrock and colleagues (2014), only 0.5% of respondents referred to asylum seekers, even though they made up about 7% of the foreign population at the point of their data collection (Destatis, 2017). Thus, it appears that salience is responsive to changes in both the composition and the portrayal of the foreign population, yet with significant inertia.

Opportunities for Framing
In light of the fluctuations, the findings highlight one way in which framing and agenda setting can be effective (cf. McCombs & Shaw, 1972). Without changing underlying attitudes, shifting who people think about might change their reported attitudes towards foreigners. Our moderation results show that negative attitudes towards Muslims are translated into negative attitudes towards foreigners particularly among those who think of Muslims when they think of foreigners. Given that anti-Muslim prejudice is stronger than most other types of prejudice in many Western countries (Heath & Richards, 2019;Statham, 2016;Strabac & Listhaug, 2008), getting people to think of Muslims when they think of foreigners might be an effective strategy for those trying to shore up negative sentiments. However, in this context, it is important to note the findings of Spruyt and van der Noll (2017) who show that chronic salience (i.e. spontaneous association by the respondent) has a stronger effect than contextual salience (i.e. a definition provided in a questionnaire). Thus, it appears that framing interventions would need to be sustained so that they shift spontaneous associations, rather than one-off attempts at priming a respondent.
It appears that this has already been recognized by right-wing populist groups who portray foreigners as an undifferentiated mass exemplified by Muslims and refugees (Mudde, 2007). However, observers often neglect to consider that the resulting support for xenophobic positions might be shallower, or at least of a different nature, than it appears, in that it does not refer to the majority of the actual foreign population. Consequently, public attitudes might be amenable to interventions that highlight the diversity of foreigners and the large share of foreigners who are less "foreign" than suggested by the prevalent attitude objects.

Implications for Research
The results clearly confirm the underlying hypothesis that the attitude objects respondents refer to when expressing attitudes towards foreigners are highly diverse, and that this diversity explains a significant share of the variance in self-reported attitudes. While a majority of respondents kept the largest group of foreigners in mind, thereby creating a widely shared prototype, more than 80% did not exclusively refer to Turkish people, which results in a wide variety of salient groups that are referenced when responding to questions about "foreigners." Such diversity may generate survey responses that differ between respondents even if their attitudes towards any one specific group are the same. This needs to be borne in mind when interpreting research findings, and depending on the purpose, measures of attitudes either towards specific sub-groups of foreigners, or towards a more explicitly defined attitude object may be preferable.
If foreigners are considered as a uniform category, and policies are responsive to research findings, they may be misguided. As Blinder (2015) reported, in response to perceived public opinion about excessive immigration, British immigration policies were tightened towards international students, who did not form part of the immigrant attitude object of most members of the public, and were thus unlikely to be central to their concerns. Similarly, research into intergroup contact and its effect on generalized attitudes might be at risk of misinterpreting the underlying processes if the generalization is caused by a shift of the attitude object rather than a change in attitudes towards groups not encountered.

Limitations and Further Research
The current dataset focused on one point in time. As the comparison with Asbrock and colleagues (2014) shows, there are trends over time, which could be profitably studied with a longitudinal approach. Likewise, our study focuses on a single country. While key findings converge with the Belgian results of Spruyt and colleagues (2016), a coherent international dataset using an open-ended question would be beneficial. This would also add to our understanding of the null finding regarding refugees. Additionally, the cross-sectional design does not allow us to rule out reverse causation in that negative attitudes towards foreigners might lead respondents to pay increased attention to groups of foreigners that are seen as particularly negative. This could be expected in line with the research into confirmation bias, where it has been shown that subjects prefer to cognitively engage with items that confirm their attitudes, be these posts on social media (Bessi et al., 2015), pieces of information bought in an experimental setting (Jones & Sugden, 2001) or published academic findings (Nelson, 2014). While our finding that the salience of Muslim groups matters even when controlling for attitudes towards them provides an indication that this is at least not the only effect at play here, further experimental or longitudinal work could profitably explore this possible mechanism.
Furthermore, the current dataset does not contain measures of individual differences that might predict group salience or moderate its links with generalized attitudes towards foreigners. One such moderator to be explored is right wing authoritarianism, which has recently been suggested to moderate the link between outgroup homogeneity and outgroup derogation (Frederic & Falomir-Pichastor, 2018). Similarly, our one-dimensional measure of attitudes towards foreigners could be complemented by multi-dimensional scales in the future to see whether the effect of group salience on prejudice differs between different aspects of prejudice. Additionally, the implications of the findings in part depend on how stable the attitude object is with regard to manipulating the perception of foreigners and to framing the elicitation of the attitude. While one-off framing has been shown to be of limited effectiveness in one study (Spruyt & van der Noll, 2017), its practical and theoretical importance suggests the need for further work. Finally, we were only able to consider two dimensions of distance; others, such as ethnicity or national income, were too closely associated with religion in our dataset to allow for meaningful comparisons. Future research should focus on the interaction between multiple types of distance to identify whether, for instance, an increased salience of gateway groups (groups close on one, yet distant on another dimension, cf. Love & Levy, 2019) might be associated with more positive attitudes, and therefore provide an avenue for interventions.

Conclusion
Studying attitudes towards foreigners and interpreting the results of any such studies requires a recognition that the underlying attitude object is complex and is typically made up of groups that are more distant from the host population than the 'average foreigner.' The first group that comes to mind appears to exert a particularly strong influence, so that, at least if that group is Muslim, all foreigners can appear to be painted with one brush. Mentioning Muslims first, or only, as a particular case of more distant groups, is associated with more negative attitudes towards all foreigners. However, this was notably not the case with regard to refugees, likely because thinking of them can make a wide variety of categories salient, and thus trigger different attitudes. In any case, the apparent diversity in attitude objects and its impact on evaluations should be considered in any research that refers to foreigners as a category, be it focused on populist movements, intergroup contact, public policy or other related fields.

Notes
i) The initial pre-registration can be found on AsPredicted.org (https://aspredicted.org/v3z7k.pdf). Based on conversations with colleagues and anonymous reviewers, we made several changes to the analytic approach, primarily by including all responses rather than just the group mentioned first and by shifting from ANOVAs towards multiple regressions that are easier to interpret. Additionally, we are not reporting the planned comparisons of two measures of anti-foreigner prejudice as our results were consistent across them and the duplication would add little insight. Finally, we reduced the dimensions of distance under consideration: originally, we wanted to test the impact of the national income level and the majority ethnicity of the salient groups of foreigners alongside their religion and refugee status, but realized that these dimensions were too closely associated in our data. We believe that these changes improved our approach and are grateful to those who provided relevant suggestions; the results presented here do not differ substantively from those initially obtained when following the approach set out in the pre-registration.

Funding
The authors have no funding to report.