Hidden Sources of Anti-Muslim Attitudes: Joint Effects of Interactions and Exposure to Out-Groups

Abstract Interactions between social identity groups can reduce perceptions of threatening out-groups and improve inter-group attitudes. But these interactions have an inevitable side effect: while an interaction may improve attitudes among its participants, the same interaction can increase exposure to out-groups in the proximity of the interaction, leading to increased perceptions of threat among those not participating in the interaction. With such negative externalities in mind, this paper argues that the presence of a large number of out-group members both improves and aggravates native attitudes toward out-groups in the same area, which may, in the aggregate, conceal a hot spot of anti-immigration attitudes. This study examines the effects of interaction and exposure through a series of surveys of native attitudes toward Muslim immigrants in the Netherlands. While the exposure effect was not observed, empirical analyses suggest that brief interactions tend to worsen negative attitudes toward Muslims, possibly due to their physical and religious appearances. This highlights the importance of visual cues in shaping inter-group relations, as these visual cues may prompt natives to sort out interactions based on appearance, hindering efforts to promote inter-group contact between Muslims and non-Muslims.


Introduction
Negative attitudes toward immigrants are common around the world. Previous scholarship has found that native populations often see immigrants as threats (Card, Dustmann, andPreston 2012, Hainmueller andHopkins 2014;Scheepers, Gijsberts, and Coender 2002;Citrin 2007a, 2007b;Sniderman and Hagendoorn 2007), with the presence of a large number of immigrants or (sudden) increases in the immigrant population fueling anti-immigrant sentiments (Burgoon and Rooduijn 2021), particularly when local residents believe their neighborhoods are directly affected (Cortina 2020;Ferwerda, Flynn, and Horiuchi 2017;Zorlu 2017). Attitudes toward immigrants can also be influenced by factors such as media attention and public discourse (Hopkins 2010), expert opinion (Kage, Rosenbluth, and Tanaka 2022), the structure of the labor market (Peters et al. 2019), and the welfare dependence and economic conditions of immigrants (Avdagic and Savage 2021) and natives (Dancygier and Donnelly 2013;Goldstein and Peters 2014;Malhotra, Margalit, and Mo 2013;Scheve and Slaughter 2001).
One well-known strategy to soften negative sentiment toward out-groups is to increase social contact between in-groups and out-groups, supported by recent experimental studies that have found inter-group contact to reduce prejudice and discrimination (Finseraas and Kotsadam 2017;Finseraas et al. 2019;Lowe 2021, Mo andConn 2018;Mousa 2020;Scacco and Warren 2018). By contrast, a growing number of studies on the related but distinct concept of exposure-being surrounded by members of the out-group through casual observation, without directly interacting with them-find that mere exposure to immigrants can negatively affect natives' attitudes (Dinesen and Sønderskov 2015;Enos 2014;Hangartner et al. 2019).
Interaction and exposure are inseparable in the real world as any interaction inevitably introduces exposure effects in its surroundings. Some studies thus examine social interactions and exposure simultaneously (Gijsberts, van der Meer, and Dagevos 2012;Vervoort, Flap, and Dagevos 2011;Zorlu 2017). Although perceptions of threat may grow with the number of immigrants (through exposure), opportunities for contact also increase, thereby leading to more positive attitudes in areas with many immigrants (Maxwell 2019;Savelkoul et al. 2010).
While this paper simultaneously considers interactions and exposure, it goes a step further by arguing that even when interactions in the aggregate have a positive impact in locations with many out-groups, they can induce negative attitudes among natives in surrounding areas by increasing exposure. More precisely, even if an interaction with an immigrant has a positive effect on a native's attitudes toward out-groups, the very same immigrant can harden the attitudes of other natives in the proximity of the interaction who do not enjoy opportunities for positive interaction. An important implication of the argument is that the same locality will harbor the strongest positive and negative attitudes toward out-groups-variation often concealed by limited data for smaller units of analysis.
Against this background, the main goal of the paper is to make a theoretical contribution to the literature by advancing our understanding of the conditions under which interactions have positive and negative effects on native attitudes toward out-groups. The paper also provides preliminary evidence for the joint effects of contact and exposure.
In my empirical analyses, I examine the effects of short interactions between Dutch natives and Muslim immigrants as well as their potential exposure effects. Muslim populations in Europe are often considered the group most likely to induce negative attitudes among native populations Valfort 2014, 2016b;Blinder, Ford, and Ivarsflaten 2019;Dancygier 2017;Helbling and Traunmüller 2020;Savelkoul et al. 2010;Sniderman and Hagendoorn 2007). 1 This creates a problem for empirical research as natives who object to accommodating Muslims in their midst may already be living far away from them-that is, the spatial distribution of attitudes toward the Muslim population may be an artifact of natives' initial preferences (see Logan and Molotch 1987;Modai-Snir and Plaut 2021;Zorlu and Lattens 2009, for discussion on residential sorting). 2 To address this problem, I administered a pre-registered, small-scale, field quasiexperiment on the outskirts of the city of Amsterdam, allowing me to systematically examine the joint effects of contact and exposure. I then conducted an online survey experiment which examined whether factors other than actual interactions could explain attitudes toward Muslims.
Contrary to my expectations, the field quasi-experiment found that short interactions did not reduce perceptions of threat. My analysis shows that when individuals directly interacted with Muslim confederates-research assistants going door-to-door with a petition unrelated to immigration-their support for outgroups decreased, while I did not detect a negative exposure effect.
A separate survey experiment investigated the reasons behind the negative attitudes observed in the interactions with the Muslim confederates. This survey experiment focused on the effect of visual characteristics and found that native Dutch respondents had strong negative reactions to the Muslim confederates' religious appearance (i.e., wearing a headscarf) and Arabic facial features. The findings suggest that the negative effect observed in the field quasi-experiment is partially driven by perceived physical appearance and that short interactions were not enough to alleviate already-held negative perception. Overall, the paper's findings indicate that short-term interactions may actually worsen negative attitudes toward Muslims due to the visual cues of their physical and religious appearances. This also highlights the challenge of promoting inter-group contact between natives and Muslims, as the importance of visual cues may prompt natives to avoid interactions with Muslims in the first place.
Research conducted in Western countries has consistently shown that Muslims face the highest levels of hostility (Adida, Lo, and Platas 2018;Adida, Laitin, and Valfort 2016b;Blinder, Ford, and Ivarsflaten 2019;Helbling and Traunmüller 2020;Savelkoul et al. 2010;Sniderman and Hagendoorn 2007). This paper indicates that overcoming entrenched negative attitudes toward Muslims may require more than short positive interactions. To improve inter-group relations, future research may thus explore a range of other strategies, such as perspective-taking (Adida, Lo, and Platas 2018), narrative persuasion, and non-judgmental listening (Broockman and Kalla 2020) as well as longer-term interactions that can lead to meaningful social connection (McLaren 2003). However, it is important to note that even if an effective interaction strategy is identified, negative exposure effects may still occur, while non-Muslims may also visually sort out interactions with Muslims. Therefore, further research is needed to better understand what kinds of interactions benefit inter-group relations in real-life settings.

Threat Perceptions and Contact
This paper argues that studying inter-group interactions and exposure to outgroups through the same analytical lens is the key to better understanding spatial variation of attitudes toward out-groups. To this end, by building upon inter-group contact theory and ethnic competition theory, this section first elaborates on the relationship between perceptions of threat (which can be associated with exposure) and contact.
According to ethnic competition theories, natives tend to perceive immigrants as threats, which generates negative attitudes toward them (Coenders et al. 2004;Scheepers, Gijsberts, and Coender 2002). The expectation is based on two mechanisms. First, limited resources induce a sense of competition between social groups, leading to mutually negative attitudes (Scheepers, Gijsberts and Coender 2002). Perceived threats are deemed to be the direct cause of negative attitudes toward minority groups (Bobo 1983). Second, human beings have a psychological tendency to cultivate negative attitudes toward out-groups (Tajfel and Turner 1979). In particular, individuals tend to perceive their own group as superior to other groups and see negative characteristics in out-groups. Combining the two mechanisms, ethnic competition theories posit that this process of social identification against out-groups intensifies when natives perceive immigrants to be threatening (possibly based on actual inter-group competition) which leads to negative attitudes.
Perceptions of threat can be reduced through inter-group contact. According to inter-group contact theory, contact between members of different groups can reduce mutually held negative attitudes under conditions such as the existence of common objectives, equal status, favorable legal provisions, and friendship (Allport 1954). A recent meta-analysis, however, suggests that these conditions are not necessary to elicit positive attitudes toward out-groups (Pettigrew and Tropp 2006). While Paluck, Green, and Green (2019) point to the lack of experimental research systematically examining scope conditions, more recent experimental research shows that short-term interactions that fall short of Allport's optimal conditions can also induce more positive attitudes toward out-groups Kalla 2016, 2020;Choi, Poertner, and Sambanis 2023;Karim 2020). For example, Broockman and Kalla (2020) find that 10-minute conversations are enough to soften people's exclusionary attitudes toward transgender people and unauthorized immigrants. 3

Impact of Short Interactions
Still, the effects of interaction are generally difficult to pinpoint in real-life settings as people self-select whom they interact with; if individuals already have negative attitudes toward specific out-groups, they will likely avoid interacting with their members in the first place (Logan and Molotch 1987;Modai-Snir and Plaut 2021;Zorlu and Lattens 2009). Such selection effects are well documented, with natives residing in large cities tending to have more pro-immigrant views than their rural counterparts (Maxwell 2019). In this context, short interactions (and exposure) are likely more relevant than full-fledged contact in real-life settings as residential (racial and ethnic) segregation continues to prevail across advanced democracies (Enos 2017;Maxwell 2019Maxwell , 2020Modai-Snir and Plaut 2021;Trounstine 2018;Zorlu and Lattens 2009). Although individuals may commute to work, research finds that mobility remains informed by race and ethnicity and that workplace segregation remains substantial (Candipan et al. 2021). This suggests that full-fledged contact is less likely than exposure and short interactions, and that even when full-fledged contact happens, it may well be initiated by people's initial willingness to interact with outgroups.
What kinds of short interactions are relevant? Broockman and Kalla (2020) studied 10-minute conversations between natives and immigrants, while Choi, Poertner, and Sambanis (2023) find that just overhearing a conversation of an immigrant can have a positive impact under some conditions. McLaren (2003) argues that natives tend to perceive out-group members to have different morals, values, beliefs, and attitudes (see also Kinder and Kam 2009) and suggests that interactions can improve attitudes if they reveal that the immigrant has similar beliefs as natives. Indeed, Choi, Poertner, and Sambanis (2023) only observed improved attitudes when conversations that natives overheard revealed that immigrants have progressive views on gender. The discussion thus far suggests that if contact involves verbal interactions that show immigrants think like natives, these interactions can soften in-group attitudes toward out-groups. Such positive interactions may also soften negative attitudes toward immigrants even when interactions are short in duration. From this, this paper derives the following baseline hypothesis: H1: Direct interaction with immigrants including verbal conversation suggesting that they think similarly improves natives' attitudes toward them.

Joint Effect of Interaction and Exposure
As indicated above, recent research shows that inter-group contact can positively affect people's attitudes toward different groups. I argue that even short-term interactions can induce such improvements. But in contrast to the laboratory experiments that social contact research relies on-Mousa (2020), Scacco and Warren (2018), and Lowe (2021) are recent exceptions-contact between social groups in the real world has an inevitable side effect: positive interaction in one location induces exposure to out-groups in the spatial proximity of the interaction.
Importantly, unlike interactions that individuals tend to self-select, exposure is more difficult to avoid. Exposure is therefore more likely to be experienced by members of diverse communities and to have greater real-life implications (Dinesen and Sønderskov 2015). A growing number of studies have found that exposure to immigrants negatively affects native attitudes (Dinesen and Sønderskov 2015;Enos 2014;Evans and Ivaldi 2021;Hangartner et al. 2019). Enos (2014)'s large-scale field experiment in the USA found that negative attitudes toward out-groups (i.e., Hispanic Americans) increased when natives were exposed to randomly assigned Hispanic confederates, suggesting that mere exposure to immigrants can aggravate native attitudes in natural settings. Relying on a natural experimental setting in the Greek Aegean Islands, Hangartner et al. (2019) found that mere exposure to refugees increased local residents' hostility toward refugees, immigrants, and Muslim populations as well as support for restrictive asylum and immigration policies.
The theoretical framework attributes this aggravation of attitudes to exposure, which can increase perceptions of threat as natives observe immigrants without having actual interactions with them. Even when native colleagues may have positive interactions with their Muslim counterparts, other natives may only observe the presence of out-group members without opportunities to interact with them. Parallel to Enos (2014)'s findings, perceptions of threat may grow among natives in the proximity of such positive interactions. In addition to the visual exposure to immigrants, natives in the proximity of interaction may also mention the (sudden) presence of Muslims in the neighborhood during casual conversation. 4 I argue that this may have a similar effect to mere observation by increasing perceptions of threat, without the opportunities of contact to soften negative attitudes.
An assumption underlying the negative effect of exposure is that natives, in the aggregate, have baseline negative attitudes toward immigrants. If this is the case, interactions can potentially flip negative attitudes while negative attitudes are only aggravated through exposure. It is of course possible that the natives who observe interactions between natives and immigrants soften their attitudes, particularly when interactions appear to be positive (i.e., potential positive externalities of exposure). Recent findings also suggest that exposure to esteemed immigrant celebrities (e.g., Mohamed Salah, an Egyptian footballer who plays for Liverpool) can reduce natives' negative attitudes toward immigrants with the same background (Alrababa'h et al. 2021). Still, there are at least two reasons to believe that the negative effects of exposure are stronger than its positive effects. First, exposure can take place without observing actual interactions. Second, while natives who have positive interactions with immigrants may share their positive experiences with neighbors, areas of interaction are geographically smaller than exposure areas (partly due to the self-selection of interactions). It is thus likely that negative exposure effects diffuse more readily than positive interaction effects.
Overall, I expect that in a society where attitudes toward immigrants are on average negative, short interactions between natives and immigrants can induce more positive attitudes among natives, but with interactions creating negative exposure effects. From this, this paper derives the spatial hypothesis regarding contact: H2: Attitudes toward out-groups become more negative among natives who are in the proximity of interaction but who do not directly interact with the same immigrants. 5 Examining the hypotheses contributes to the literature. Recent macro-level studies suggest that although perceptions of threat may heighten with the number of immigrants, opportunities for contact also increase, leading to more positive attitudes in areas accommodating many immigrants (Maxwell 2019;Savelkoul et al. 2010). A key assumption is that contact between natives and immigrants occurs more often in these areas.
This paper challenges the prevailing assumption about interaction opportunities by arguing that they are not evenly distributed within localities, with some natives deprived of opportunities to interact with immigrants or actively seeking to avoid them. While natives living in large cities may on average have more interaction opportunities than natives living in rural areas, there still remains important spatial variation within localities in the extent of interaction opportunities (see also Musterd 2005). This is partly because immigrants tend to live and work in spatially segregated areas (see also Enos 2017; Maxwell 2019; Trounstine 2018), partly because natives vary in their choice of activities that facilitate interaction with immigrants (Candipan et al. 2021;Feld 1981), and partly because some natives avoid interacting with immigrants.
It follows that even in the same city, some natives have many interaction opportunities while others just observe the presence of a large immigrant population. I then argue that the presence of immigrants has a positive effect on natives who have direct interactions with them, but that the same immigrants have a negative impact on natives who live and work in the vicinity and only experience exposure.
The study has implications for the literature on attitudes toward immigrants. If areas with more immigrants provide more interaction opportunities, I expect, on average, more positive attitudes toward immigrants there. But I also expect the strongest negative attitudes in such localities due to increased exposure and perceptions of threat stemming from large immigrant populations in parts of the same area. As research tends to aggregate individual attitudes within larger units such as municipalities or the European Union's NUTS-3 regions (average population 150,000-800,000), scholars are likely to miss important variation within larger units (see also Dinesen and Sønderskov 2015;Sluiter, Tolsma, and Scheepers 2015). Existing data from the Netherlands confirm this intuition. Using observational data (the Social and Cultural Developments in the Netherlands (SOCON) survey), Figure A.1 in the Appendix shows that while the three largest Dutch cities with the most immigrants-Amsterdam, The Hague, and Rotterdamcontain large number of natives who do not see immigrants as threats, the three cities also include natives who most strongly perceive minority groups as threatening. Figure A.2 in the Appendix also shows that attitudes toward immigrants in other cities are more uniformly distributed. Although this analysis is exploratory and cannot establish causal relationships, the figures indicate that the presence of a large number of immigrants may both improve and aggravate native attitudes toward them in the same area, which may, in the aggregate, conceal a hot spot of anti-immigration attitudes. Therefore, further research is needed to test the joint effects of contact and exposure more precisely, using smaller units of analysis. This motivates the field quasi-experiment below.

Research Design
Case Selection I conducted a series of survey and field quasi-experiments to test the implications of the above-stated hypotheses about the joint effects of interaction and exposure. The Netherlands is geographically among the smallest of democracies, well-suited to examine my argument on micro-level interactions between members of in-and outgroups. Like many other European countries, there is significant anti-Muslim sentiment in the country (Savelkoul et al. 2010;Sniderman and Hagendoorn 2007). According to polls conducted in March 2018, 38% of Dutch citizens identified migration and integration as the most pressing national issue (Ridder, Dekker, and Boonstoppel 2018), and among various out-groups, Muslims are often regarded as the "unacceptable other" (de Koning 2016) (see also below for a more general discussion about Muslims in Europe).
The surveys also took place in 2018, with the 2015 immigration crises in Europe fresh in people's memories. Thus, I expect Dutch natives to have, on average, negative attitudes toward Muslim immigrants, regardless of location. That is, people should tend to react to the presence of Muslim immigrants (i.e., exposure) negatively.
Still, like other European countries, the Dutch government adopts the so-called citizenship approach and encourages the integration of ethnic minorities into Dutch society (Poppelaars and Scholten 2008)-in line with an underlying condition of the contact hypothesis (Allport 1954). While Dutch natives are generally considered to be tolerant toward minorities, there is also a strong expectation that immigrants conform to Dutch values and standards of behavior (Gordijn 2010). This has important implications for the scope conditions and research design of this paper. If Dutch people are on average quite tolerant but care a great deal about conformity to social norms, I expect that Dutch natives, despite their underlying negative attitudes at the time of the surveys (see also Appendix A9), can cultivate positive attitudes toward Muslims when they interact with them and perceive that Muslims share similar ideas and speak and behave like them, that is, when they perceive that Muslims are willing to conform to Dutch society and culture.

Empirical Strategy
This paper's empirical strategy entails two steps. First, I present findings from a field quasi-experiment that examines the microfoundations of my hypotheses. More concretely, the experiment examines whether actual interactions with Muslims would overcome perceptions of threat and soften Dutch natives' attitudes, while increasing threat perceptions in the proximity. The field quasi-experiment assigned two groups-Dutch native confederates and Muslim confederates-to two similar localities. The intervention was accompanied by a two-wave survey before and after the intervention. Second, I present results for an additional online survey experiment that tests which aspects of interactions with the Muslim confederates elicited negative attitudes in the field quasi-experiment (see below for more detail on each study's design).
I also conducted another survey experiment to gauge Dutch natives' baseline attitudes toward Muslims. The experiment examined whether a hypothetical relocation of Muslim refugees elicits negative attitudes among Dutch natives and how distance from the relocation point can affect their attitudes. The results suggest that Dutch people are, on average, reluctant to interact with Muslims, and consistent with the not-in-my-backyard (NIMBY) literature (e.g., Ferwerda, Flynn, and Horiuchi 2017), they do not want out-groups in your proximity. This confirms the paper's assumption that Dutch natives generally have negative attitudes toward Muslims and are often reluctant to interact with them (more detailed research design and results can be found in Appendix A9).

Field Quasi-Experiment
I administered the small-scale, field quasi-experiment between May and July 2018. The experiment was pre-registered at Open Science Framework. In the experiment, I recruited two groups of female research assistants from a Dutch university who served as confederates. 6 They were instructed to collect signatures for a petition for a neutral cause (see below) in the streets of a village right outside the city of Amsterdam where the majority of residents are native Dutch. The village is quite wealthy compared to other parts of the Netherlands. Without concerns about economic competition (Gerber et al. 2017), I expected the native residents of the village to be more likely to improve their attitudes toward out-groups through interaction with Muslim confederates.
The two groups of confederates-one consisting of two White individuals and the other of two individuals with Middle Eastern backgrounds wearing headscarveswere placed in two different but comparable streets in the same village for a few hours in the afternoon on two consecutive weekends. The streets were approximately 0.5-1 km apart. After arriving at pre-assigned households, the confederates rang the doorbells, explained their cause, and collected signatures for their petition. More specifically, they told residents that Dutch universities are dominated by students from cities, leaving rural areas under-represented, and that they were concerned that this limited social (not ethnic or religious) diversity may have broader implications for Dutch society. The main goal was to induce the idea that both groups of confederates care about the country, that they speak and behave like other Dutch people, and to cultivate positive attitudes toward Muslims. Although it is possible that some local residents disagree with the cause of the petition, they should also understand that the confederates care about the country.
All the confederates followed a strict protocol when introducing themselves and were instructed to follow the natural flow of the conversation depending on what residents brought up. Each group visited about 60 households and revisited the same streets in consecutive weeks to maximize coverage. 7 To ensure that the presence of the confederates was noticed by neighbors (i.e., exposure effect), they also spent time in the streets where they collected signatures before and after the campaign (see Section A3 in the Appendix for more detail about the intervention).
The field intervention with the two groups of confederates created two different treatment arms: (1) White confederates vs. Muslim confederates and (2) direct interaction vs. no direct interaction. The design yields four conditions: (1) exposure to White confederates; (2) exposure to Muslim confederates; (3) interaction with White confederates; and (4) interaction with Muslim confederates. Since this paper is interested in both the effects of interaction and exposure, I treat the first condition (i.e., exposure to White confederates) as the control group; since the White confederates look like other native Dutch residents, they were likely to be considered passersby. If my expectation is correct, I should see that those in the direct interaction area with the Muslim confederates softened their attitudes toward Muslims, while those in the no-direct interaction area with the Muslim confederates hardened their attitudes.
The intervention was not randomly assigned as the study only has two main treatment groups (i.e., interaction with White confederates and interaction with Muslim confederates) and it contacted all households on each street. The intervention is thus a field quasi-experiment. The main concern for the non-random treatment is that it may generate imbalances in covariates between the groups. 8 To ensure that the observable covariates between the groups are comparable, this paper adopted a twowave survey and sought to control for differences in observed factors between the two groups. Table A1 reports summary statistics for the four treatment groups.
The surveys were conducted before and after the intervention to examine the intervention's effect on residents' attitudes. The surveys were constructed broadly and included many questions unrelated to immigration to lessen their connection to the field intervention. Households being the unit of analysis, I sent an invitation via mail to participate in the survey to each household in the village I could identify through publicly available information (N 2; 965). 9 As Stedman et al. (2000) finds, response rates for mail survey are generally low and the rates have been declining over time. But due to budget constraints, I chose the mail survey over faceto-face surveys.
Survey responses were collected through the online survey platform Qualtrics. Respondents received invitation letters by mail with links to the survey and a unique registration code that enabled linking the two survey waves and the conditions to respondents. To encourage participation, I attached a small gift (a pen) to the letter, which stated that the study sought to understand which aspects of the residential area were important to residents. The surveys took about 10 minutes to complete. I collected 247 responses in the pre-intervention survey. Of these respondents, 137 participated in the post-intervention survey, which is the total sample size. Table A.2 reports attrition rates, which are comparable across the treatment groups except for "Interaction with Muslim confederates," where I found residents to have lower attrition rates than for the other three groups.
The main dependent variable is Attitudes toward Muslims, measured both in preand post-intervention surveys. 10 The variable is based on responses on a 5-point Likert scale (1 = fully agree, 5 = fully disagree) to 13 items, taken from the Social and Cultural Developments in the Netherlands (SOCON) survey, which have been used in previous studies (e.g., Savelkoul et al. 2010). Sample items read "Muslims who wear a headscarf do not adjust to our society" and "Muslims raise their children in an authoritarian way" (see Section A2 in the Appendix for all items). I averaged across responses to obtain the measure of attitudes toward Muslims. The Cronchbach's alpha was 0.91 in the pre-intervention measure and 0.93 in the post-intervention measure. Higher scores mean more positive attitudes toward Muslims. Table A.3 summarizes the dependent variables.
To control for differences between groups in the intervention, this paper took advantage of the two-wave panel survey and estimated the impact of interaction as well as the spatial impact of interaction on attitudes toward Muslims by using a difference-in-difference framework. This means that all the treatment variables are interacted with a time variable (i.e., Wave 2).
Following similar studies (Adida, Lo, and Platas 2018; Broockman and Kalla 2020; Savelkoul et al. 2010;Sniderman, Hagendoorn, and Prior 2004), I performed ordinary least square regression analyses 11 and clustered standard errors on the level of participants. I also included pre-registered treatment covariates to adjust remaining imbalances across different treatment groups (see also Gerber and Green 2012). In all analyses, I controlled for respondent gender, age (in years), education, ethnicity, household income, duration of residence, and spatial distance to the intervention point (km). Specifically, point estimates and confidence intervals are calculated from the following regression form: Y it α βCondition i γWave t δ Condition i Wave t λCovariate i ε it , where Condition i is the treatment group-specific effect, Wave t is the time trend common to the control and treatment groups, and Covariate i is a vector of control variables. A full regression table is reported in Table A.4 in the Appendix. 12 Following Brambor, Clark, and Golder (2006), I illustrate marginal effects of treatments in Fig. 1 (see Figure A.4 for the marginal effects with standard errors). Figure 1 indicates that respondents who did not interact with the confederates did not experience any significant changes in their attitudes over time ("Indirect contact White confederates" and "Indirect contact Muslim confederates" in the figure). Accordingly, I did not find the sudden presence of Muslims to have any significant impact on residents in the vicinity of the interaction (β 0:02, SE 0:09, p 0:78), which is not consistent with H2.
Importantly, Fig. 1, along with the top panel of Figure A.4 and Table A.4, shows that relative to the pre-intervention measure, in the post-intervention measure, those who had direct interaction with the Muslim confederates reported significantly less positive attitudes toward Muslims.
It is possible that residents who interacted with the Muslim confederates already had hardened attitudes toward Muslims, and that their attitudes had little to do with the interaction. But comparing the trends in the attitudes of residents in the same area (i.e., those who interacted with the Muslim confederates and those who live in the surroundings), this explanation may be unlikely. Participation rates may also have affected the results. The rates varied between the treatment groups: 30 out of 62 households (48.3%) in the Muslim confederate street opened the door, compared to 40 out of 57 households (70.1%) in the White confederate street. Lower participation in the Muslim confederate street could have been driven by residents' pre-existing antipathy toward Muslims. 13 If this is the case, the negative interaction effect the study found could well be under-estimated. 14 Still, this is only speculation, and it is also possible that the residents were simply not present at the time of the confederates' visits.
Interestingly, the figure also shows that interactions with White confederates had a negative impact on residents' attitudes toward Muslims. This negative effect may have been attributed to residents perceiving university students as more liberal (Thijs, Grotenhuis, and Scheepers 2018), with the confederates' promotion of a petition on social diversity possibly causing annoyance and a backlash. While this remains speculative and the effect is not statistically significant at the 10% level, the results suggest that discussions on diversity alone may have a negative impact on attitudes toward Muslims, even in the absence of direct interactions or exposure to Muslims. This, in turn, highlights widespread hostility toward Muslims in Dutch society and the need for further research to better understand the underlying factors contributing to this phenomenon.
As for other covariates, Table A.4 in the Appendix that includes the findings for the covariates shows that all else equal, more-educated respondents are more likely to exhibit positive attitudes toward Muslims than less-educated counterparts, which is consistent with general findings about the relationship between education and attitudes toward immigrants (Hainmueller and Hopkins 2014;Hainmueller and Hiscox 2007). The table also shows that non-binary respondents are more likely to be supportive of Muslims than male counterparts. This is somewhat surprising in light of previous studies on the conflicting relationship between Islam and its perceived gender views (Eskelinen and Verkuyten 2020;Kwon, Scarborough, and Taylor 2023). However, it is consistent with previous research showing that nonbinary individuals tend to be politically liberal (Worthen 2020).
One caveat is that the field quasi-experiment is under-powered due to a number of non-responses. Although the other studies (the online survey below and another survey in the Appendix) compensate for this limitation and provide important findings, there is a need for further research to systematically investigate the joint effect of interaction and exposure.
The next study aims to investigate why residents' attitudes were negatively impacted by interactions with Muslim confederates in the field quasi-experiment. According to the confederates as well as an independent observer (another research assistant who monitored the door-to-door visits), interactions with residents were equally positive and smooth. The residents, however, may have judged their interactions with the Muslim confederates more negatively due to their physical appearance. Alongside physical markers of race and ethnicity and the presence or absence of a headscarf, the confederates may have had other physical characteristics less related to ethnicity or race but commonly perceived as signaling kindness and trustworthiness, for example, neoteny or facial attractiveness. To examine these, the study conducted a separate online survey.

An Online Survey
To investigate the appearance effect, I took photos of the confederates' faces and had them rated by an independent sample of Dutch residents in December 2018. The sample was drawn from an online panel of Dynata. 15 The number of observations was 1,810. 16 All pictures were standardized for background, lighting, angle, distance, and facial expression. The photos were cropped so that only the face and neck were visible and were aligned so that the eyes were centered. For each confederate, I took three pictures: one with a headscarf and a neutral expression, one without a headscarf and a neutral expression, and one with a headscarf and a smiling face. 17 The literature generally suggests that physical appearance can influence how individuals are treated (Hamermesh 2011), and this can also be true in inter-group relations (Kleider-Offutt, Bond, and Hegerty 2017;Maddox and Perry 2017). Moreover, studies have highlighted that racial phenotypicality, which refers to observable physical traits associated with a person's race, can have an impact on how individuals are evaluated. For instance, White Americans are more likely to negatively evaluate Black Americans with more Afrocentric facial features (Blair, Judd, and Fallman 2004), while young Black men are perceived as more physically threatening than young White men (Wilson, Hugenberg, and Rule 2017).
Negative treatment toward out-groups can often stem from implicit negative attitudes toward the physical appearances of non-White individuals, as in many cultures, physical features associated with White Eurocentric phenotypic characteristics are considered more desirable (Maddox 2004). These preferences for certain facial characteristics can then influence how individuals are categorized and perceived, ultimately impacting social interactions and group dynamics (Maddox 2004). In the context of this paper, it is thus possible that Muslim confederates with Arabic facial features may be perceived as out-group members in Dutch society because they lack typical White Eurocentric phenotypic characteristics. This can result in the confederates being subject to existing biases associated with out-group membership by Dutch natives.
The wearing of a headscarf is another notable appearance feature among many Muslim women. 18 Attitudes toward the hijab are complex and multifaceted, with some viewing it as a symbol of religious freedom (Howard 2007), 19 while others see it as incompatible with Western liberal values (Foner and Alba 2008). In Europe, many people have generally positive attitudes toward Muslims, but non-Muslims often have concerns about the role of religion in society, particularly regarding the hijab, which is often viewed as a symbol of illiberal values and as a sign of women's unwillingness to integrate (der Noll et al. 2018;Helbling 2014). In fact, a survey by Helbling (2014) reported that about one-quarter of respondents were opposed to Muslims, while almost 60% disagreed with the practice of women wearing a headscarf. These findings align with previous research indicating that in the Netherlands, many natives express more opposition to Muslim cultural norms than Muslims themselves (Sniderman and Hagendoorn (2007); see also de Koning (2016)). As such, within the context of this paper, it is possible that Dutch natives may perceive Muslim confederates wearing a headscarf negatively, as the hijab is viewed by some as a symbol of embracing illiberal values and a lack of willingness to integrate into Dutch society.
In the online experiment, I investigate the impact of both physical features simultaneously-whether wearing a headscarf and/or having Arabic facial features result in negative evaluations. A study conducted by Adida, Laitin, and Valfort (2016a) in France also examines the impact of religious symbols (on job applications) while controlling for other physical features. However, their study did not specifically examine the impact of wearing a headscarf and its relative importance compared to other physical features. As discussed earlier, it is possible that both physical and religious appearances of the Muslim confederates may affect evaluations of the confederates. But in line with the Dutch "myth" of color-blindness (see Rose 2022, for counter-evidence based on her own experience), it is also possible that neither appearance feature is important, and the negative evaluations observed in the field quasi-experiment may be attributed to other factors.
With the motivation in mind, I randomly showed respondents one of the following three kinds of photos: (1) photos with a headscarf, without a smile; (2) photos without a headscarf, without a smile; and (3) photos with a headscarf, with a smile. I showed each respondent four photos of two White confederates and two Muslim confederates. I also randomized the order of the four photos to avoid an order effect. The main design is summarized in Table A.8 in the Appendix. With this design, I sought to estimate (1) the impact of ethnic appearance (i.e., White vs. Arabic); (2) the impact of religious appearance (i.e., with or without a headscarf); and (3) the impact of positive impression (i.e., with or without a smile). Figure 2 shows a sample of (modified) photos that I presented to respondents.
Facial ratings served as dependent variables. Upon seeing photos of the confederates' faces, survey respondents rated to what extent they perceived the confederates as (1) being friendly, (2) reliable, and (3) belonging to Dutch culture on a 5-point Likert scale. Higher values indicate that confederates were rated as friendlier, more reliable, and more belonging in the Netherlands. After rating the faces and clicking the "next" button, respondents read the following text: "The girl in the photo is a promising foreign student at a university in the Netherlands and would like to stay in the Netherlands. If it was up to you, would you grant her a permanent residence permit?" (1 = absolutely no, 5 = absolutely yes). I use these four questions as the dependent variables (Likability; Reliability; Fitting-in; and Admission preference).
I conducted four separate regression analyses for the dependent variables. Standard errors were clustered among respondents. This is because each respondent rated four pictures, meaning the data are nested. As in the field quasi-experiment, I used linear regressions including covariates. 20 The results are presented in Fig. 3 The impacts of appearances on integration issues For friendliness and reliability (i.e., personal impression), the upper panel of the figure shows that compared to White faces, Arabic faces were rated as friendly but not reliable. Wearing a headscarf does not affect Dutch respondents' ratings of friendliness and reliability. Overall, the religious and ethnic characteristics of the confederates show either null or mixed findings at best. Unsurprisingly, (compared to neutral faces) smiling increased respondents' ratings positively both in terms of friendliness and reliability, the coefficient being the largest for this variable.
Regarding integration issues, this paper finds a gloomy picture for the Muslim confederates and for the Muslim population in the Netherlands more generally. The bottom panel of the figure shows that both Arabic and Muslim appearances consistently led to negative ratings. More specifically, native Dutch respondents were more likely to judge that they do not fit in the Netherlands when their face looked Arabic and they wore a headscarf. Smiling faces were still rated significantly higher for fitting in the Netherlands. Finally, respondents were less likely to grant a permanent residence permit when the confederates looked Arabic. The negative effect was even stronger for those wearing a headscarf. Unsurprisingly, the study found that smiling does not generate a residence permit.
In summary, the online survey finds that when it comes to integration issues, the Muslim confederates were more likely to be punished simply because of their ethnic (visual) background or religious symbols. The negative impact of wearing a headscarf on evaluations was greater than that of having Arabic facial features, possibly due to more explicit associations between the hijab and illiberal values (see also der Noll et al. 2018;Helbling 2014). Previous research has also shown that people tend to avoid interacting with those who look visibly different (Dietrich and Sands 2023;Walker and Hewstone 2006), which, when combined with the findings of this paper's empirical analyses, highlights the challenges in promoting social interactions between Muslims and non-Muslims. 22 While the role of visual cues may lead non-Muslims to avoid interactions with Muslims, the field quasi-experiment above also suggests that even if interactions occur, short-term positive interactions alone are insufficient in overcoming ingrained prejudice associated with Muslims. Therefore, it is crucial to develop more effective strategies to improve inter-group relations. The concluding section will discuss the policy implications of the findings.

Conclusion
In our era of global migration, social scientists have pondered how to best reduce perceptions of threat many natives attach to new immigrants and refugees. While much research suggests that interactions between natives and immigrants can ease the challenges of integration, we need to know more about whether the alleged benefits of interaction also apply to Muslim immigrants in western countries and whether inter-group interaction can produce negative externalities in the real world. This paper drew on the recent literature on social contact to argue that short interactions with Muslim immigrants can lead natives to believe that immigrants are like them, which would soften their attitudes toward out-groups. I then argued that the exclusion of natives from the inclusive bubble in the area surrounding the same interactions may in fact harden their attitudes. The implication is that there will be both strong pro-immigrant and anti-immigrant attitudes in places where many immigrants reside and work.
I tested the hypothesis of interaction and exposure in the Netherlands where natives are considered to be tolerant toward minority groups. Contrary to the predictions, the studies showed that interacting with Muslims did not lessen native perceptions of threat but instead aggravated them. Importantly, the hardening of attitudes toward out-groups was more severe among natives who had direct interaction with Muslims than among natives who did not have direct interaction but resided in the proximity of the interaction. This finding is not consistent with the hypotheses and prompts researchers to re-consider the conditions under which interactions indeed reduce perceptions of threat.
The second survey that focused on visual characteristics revealed that having Arabic facial features and religious symbols results in negative evaluations of Muslim immigrants. The findings suggest that the negative effect observed in the field quasi-experiment is, to some extent, driven by perceived physical appearance and that short interactions were not enough to alleviate already-held negative perception. In summary, the paper's findings highlight the potential of short-term interactions to exacerbate negative attitudes toward Muslims, given the salience of their ethnic and religious appearances. The findings also underscore the challenges in promoting inter-group contact between natives and Muslims, as visual cues may lead natives to selectively avoid interactions. These results emphasize the need for more effective integration tools to mitigate negative attitudes toward Muslims and foster meaningful inter-group interactions.
Muslims in Europe are often perceived as a threat (Adida, Laitin, and Valfort 2016b;Blinder, Ford, and Ivarsflaten 2019;Dancygier 2017;Savelkoul et al. 2010;Sniderman and Hagendoorn 2007), and research has shown that Muslim immigrants in Europe are often perceived to have differing political views from the native population (Choi, Poertner, and Sambanis 2023;Helbling 2014). Therefore, it may be more effective to demonstrate not just that Muslim immigrants care about Dutch society but also that they have similar political beliefs and care about Dutch society politically in the same way that Dutch natives do. Future research could thus explore a range of strategies, such as perspective-taking (Adida, Lo, and Platas 2018), narrative persuasion, and non-judgmental listening (Broockman and Kalla 2020), to improve mutual understanding about political views. However, it is important to note that even if an effective interaction strategy is identified, negative exposure effects may still occur, and natives may also visually sort out interactions. Therefore, further research is necessary to better understand what types of inter-group interactions benefit inter-group relations in real-life settings.
Supplementary material. The supplementary material for this article can be found at https://doi.org/ 10. 1017/rep.2023.15 grateful to the editor and the three anonymous reviewers for their careful reading of the paper and their comments. I also thank Marina Tulin who contributed to the project from the beginning; this research would not have been possible without her. Laurien Bender, Senja Borgers, Marly Bos, Robin Dullaert, Hayet Elabed, Lisa Lemmen, Sarah Motlagh, Philip Rakhou, Wies van der Stroom, and Anne-Sophie van Gulpen provided excellent research assistance. I also thank the University of Amsterdam's Political Economy and Transnational Governance (PETGOV) program for generous support of this project. The human subject protocol of the research was evaluated and approved by the Ethics Advisory Board, Amsterdam Institute for Social Science Research, University of Amsterdam, and the research was conducted following the approval.
Competing interests. The author declares there is no conflict of interest.

Notes
1 For the purposes of this paper, the terms Muslims and Muslim immigrants are used interchangeably, in line with the existing literature (Helbling 2014;Huijnk, Dagevos, and Vermeulen 2022). However, it is important to note that not all Muslims are immigrants. Nonetheless, Muslims in general face discrimination in Europe, and many immigrant-origin Muslims also experience challenges when attempting to integrate into society (see also Leszczensky, Maxwell, and Bleich 2020).
2 Some Muslims may also choose not to live in close proximity to native populations due to policies that limit religious expression and promote secularism and Western values, as well as perceived discrimination (see also Abdelgadir and Fouka 2020;Adida, Laitin, and Valfort 2014).
3 Although their study focuses on the effects of short-term discussions about out-groups, they find some evidence that short conversations can have a positive effect even when the canvassers are immigrants. 4 Neighbors often talk about out-groups and this can exacerbate negative attitude toward the out-groups (see also Bosson et al. 2006;Wert and Salovey 2004). 5 A similar claim has been made in electoral studies, where it is known as the "halo effect" for radical right supporters (Evans and Ivaldi 2021;Rydgren and Ruth 2013). The joint effect of contact and exposure has not been examined in experimental settings. 6 To exclude potential gender bias, I only hired female confederates. They were paid at an hourly rate for their time. 7 More precisely, 62 households for the Muslim confederates and 57 households for the White confederates. The differences are purely due to the geographical layout of the streets. The confederates visited the streets twice to reach households that were absent during the first week, while also enhancing the effect of exposure. 8 While field quasi-experiments and field experiments are comparable in many ways, for field quasiexperiments, one cannot rule out the existence of systematic differences between groups as a confounding factor (Díaz, Jiménez-Buedo, and Teira 2015). 9 The first survey was administered at the end of May 2018. The first intervention was about 2 weeks after the first wave; the second intervention was 1 week later. I fielded the second wave survey about 1 week after the second intervention. 10 I also collected data on other dependent variables such as perceived neighborhood cohesion (see, e.g., van der Meer and Tolsma 2003, for the relationship between ethnic diversity and social cohesion). The result is reported in Table A.7. 11 Figure A.3 confirms that the dependent variable is normally distributed. 12 I also conducted bootstrapping but found similar results across different bootstrap samples. 13 Residents may have been able to identify visitors through windows or a peephole. In Table A.5, I control for respondents' initial attitudes towards Muslims and confirm that the main results hold. Table A.6 also shows that having friends with minority backgrounds does not affect the results. 14 I examined the same model with another dependent variable Social cohesion and found a similar result (Table A.7). This provides additional evidence that the interaction with Muslim confederates had a negative impact. Inclusion of the Social cohesion variable on the right side in the main specification does not alter the main findings. 15 The sample stratification is based on income, residential location, education level, and gender. 16 As explained below, each of the 452 respondents was shown four different photos. I dropped non-Dutch and non-White respondents from my analysis.
17 See Horiuchi, Komatsu, and Nakaya (2012) for the effect of smiling on voters. 18 According to a recent survey, a large proportion of Muslim women in the Netherlands wear a headscarf. However, there is considerable variation within the Muslim community, with 90% of Somali Muslims wearing a headscarf, compared to only 19% of Surinamese Muslims (Huijnk, Dagevos, and Vermeulen 2022). 19 Although Muslim immigrants in contemporary Europe generally receive the most hostility from native populations in the aggregate, members of Christian minorities may, for example, see Muslim immigrants as potential allies against the threat of secularism, while many politically liberal natives are critical of Muslim immigrants for their allegedly "traditional" views on, for instance, gender roles (Carol, Helbling, and Michalowski 2015;Helbling and Traunmüller 2020). 20 Figure A.5 confirms that all dependent variables are not skewed. 21 I provide results that control for the order effect in Figure A.6. The results largely remain similar. Figure  A.7 shows the results with only the first photo seen by respondents. The direction of the results remains the same, although due to the smaller sample size, some variables lose statistical significance at the 10% level. Figure A.8 shows estimates with ordered logit, but note that some variables violate the proportional odds assumption (except the first model).
22 The paper's findings about entrenched negative attitudes toward Muslims in the Netherlands align with recent research indicating that Muslims in the country, particularly those of Turkish and Moroccan descent, face discrimination and perceive negative attitudes toward them. However, it is important to note that many Muslims in the Netherlands, especially those who were born and raised there and consider it their home (Huijnk, Dagevos, and Vermeulen 2022).