Polarization in attitudes towards refugees and migrants in the Netherlands

This research investigated whether and to what extent Dutch society is polarized in its attitudes towards refugees and migrants. We further investigated what factors were linked to polarization by testing an integrated three- dimensional model. A latent profile analysis of a representative Dutch sample ( N = 1897) suggested that society is indeed polarized, with two substantial groups— anti (16.5%) and pro (18.7%)— at opposite ends. The middle was also divided into people with critical (31.8%) and lenient (33%) attitudes. The three dimensions of our model, (1) the individual and social self (education, political orientation, relative deprivation), (2) perceptions and experience regarding refugees and migrants (perceived cultural distance and contact), and (3) the societal context (societal discontent), were all significantly associated with polarized attitudes. Interestingly, perceived cultural distance from Islam was most strongly linked to polarized attitudes. Organizing different factors into an integrated model offers new insights into the complexity of polarization in society.


| INTRODUC TI ON
The number of people forced or choosing to leave their home countries has increased dramatically over the past decade, reaching a record high of 70.8 million refugees and 258 million migrants (UNHCR, 2016(UNHCR, , 2019United Nations, 2017. Migration has become one of the greatest concerns of European citizens (CBS, 2019;European Social Survey, 2016;International Organization for Migration, 2018). While some people are willing to welcome refugees and migrants, others are reluctant to accept them as residents, negative attitudes towards refugees and migrants persist, and migration is frequently the topic of conflictual public debates (European Social Survey, 2017).
The diversity of responses towards refugees and migrants suggests that the public debate may be polarized. Polarization can briefly be described as the presence of separated groups in society with opposing attitudes (DiMaggio et al., 1996;Jung et al., 2019). Polarization may have both positive and negative outcomes on societies; it can encourage social change by stimulating discussion, political engagement, and protest (Abramowitz & Saunders, 2008;McGarty et al., 2009), but it can also undermine social cohesion by increasing the avoidance of conflicting opinions and information, the stigmatization of (minority) groups, and segregation in society (Frimer et al., 2017;McCoy et al., 2018;Moulaert et al., 2003), which may further increase attitudinal polarization. Given its potential serious impact on citizens in societies, it is important to understand the extent of attitudinal polarization within societies and the psychological factors that may explain it.
Although previous research investigated attitudes towards refugees and migrants (for a review, see Esses et al., 2017), to the best of our knowledge, there is no comprehensive understanding of the extent of attitudinal polarization in society and the psychological factors that may explain it. Therefore, we developed and tested an integrated model of attitudinal polarization in society within the context of migration. To explain polarization in attitudes towards refugees and migrants, we organized factors indicated by the social psychological literature along three dimensions: (1) the individual and social self; (2) perceptions and experience regarding refugees and migrants; and (3) the societal context. Using data from a representative sample of the Dutch population, we first examined whether and to what extent Dutch society is polarized in its attitudes towards refugees and migrants. Second, we tested our model and investigated who may hold polarized attitudes and what psychological factors might be associated with these attitudes.

| Polarization in Dutch society
Historically, the Netherlands has been known as a tolerant, welcoming country for minorities. For example, in the 17th century, the Netherlands was a popular asylum for religious refugees due to its relatively high religious tolerance (Carol & Koopmans, 2013); more recently, in 2001, it became the first country to legalize same-sex marriage (CNN, 2001). The Netherlands has also been known for its multicultural migration policies and its focus on culturally specific groups (e.g., Vasta, 2007). However, the Netherlands is currently one of a few European countries to have cut back on its established multicultural policy in favour of a stronger focus on assimilation (Banting & Kymlicka, 2013). After the increase in migration in 2015, Dutch citizens increasingly voiced concerns about this issue and criticized the government's implementation of migration policies SCP, 2016). Many Dutch people nowadays believe differences in opinions on societal issues are increasing and experience growing pressure to choose sides in the migration debate (SCP, 2019). A recent representative survey showed that although a majority of Dutch citizens believe that refugees should be granted asylum, a majority is also concerned about the consequences of such a step (Kuppens et al., 2020). Given these changes, we wondered what attitudes people in the Netherlands currently hold towards refugees and migrants.
Although a popular topic of discussion, polarization is often not well defined, and when it is defined, these definitions differ according to discipline (e.g., Jung et al., 2019). In the current research, we focus on attitudinal polarization in society and define polarization as the existence of separated groups in society, which are divided and disagree in terms of their attitudes towards refugees and migrants (for a similar approach, see Bauer, 2019;DiMaggio et al., 1996;Maher et al., 2020;McCoy et al., 2018). If there is no polarization, all individuals fall into one group; while disagreement may exist, it is not possible to identify structural groups showing increased attitudinal coherence with some individuals while simultaneously showing increased distance from others (Bauer, 2019). If there is maximum polarization, all individuals can be divided into one of two groups with attitudes at the opposite ends of the attitudinal scale. When evaluating and assessing polarization and its role in society, it is important to consider not only the presence of separated, structural groups, but also the meaning of the distances between attitudinal groups and the degree of their disagreement (Bauer, 2019). Importantly, polarization can be both a state and a process over time (DiMaggio et al., 1996); in the current research, we focus on polarization as a state.
Polarization in society is a complex phenomenon that may not be fully captured by a single observation. Therefore, similar to previous research on polarization (e.g., DiMaggio et al., 1996), we have taken a multivariate approach and assessed attitudes towards refugees, migrants, and migration in general. Although previous research has indicated that host-society members do hold differentiated attitudes towards specific refugee or migrant groups and on specific migration policy issues, there also is evidence that these attitudes are strongly correlated, thereby providing justification for the current approach in which attitudes towards refugees, migrants, and migration are measured as a general pattern (e.g., see Lee & Fiske, 2006;Postmes et al., 2017;Verkuyten, 2004).

| An integrated three-dimensional model of polarization in attitudes towards refugees and migrants
Polarization in society is a social phenomenon that operates at different levels (individual, group, and societal) and may be explained by a range of different factors. To further our understanding of polarized attitudes towards refugees and migrants, we consider it useful to approach this phenomenon as an intergroup process, and we propose a model of attitudinal polarization in which we integrate and organize different lines of literature into three dimensions. Our model specifically applies to polarization in society regarding attitudes towards an outgroup (or outgroups), whereby the current research focuses on the intergroup relationship between host-society members and refugees and migrants. As nationality or cultural background is an important way in which people categorize themselves and others (Bain et al., 2009;Prentice & Miller, 2007), we assume that members of the host-society (perhaps unknowingly or unintentionally) tend to see themselves as ingroup members, and refugees and migrants as outgroup members. Within this intergroup relation, we suggest that there are three important dimensions at play that may help explain attitudinal polarization (see Figure 1). Below, we discuss each dimension and identify specific factors within these dimensions, based on previous literature.

| The individual and social self
Individual characteristics and group affiliations are considered important factors in attitudinal polarization. The related question is: How am I and my group doing? People's perceptions of themselves and the groups they belong to may influence how they relate to and evaluate others. How people perceive their own position and that of their ingroup in society can influence how accepting they are of outgroups, and especially of refugees and migrants (Esses et al., 2001). In particular, perceptions of how secure their social position is may influence the extent to which other groups are valued or perceived as threatening. For example, host-society members may view refugees and migrants as competing for resources such as welfare and employment opportunities, thereby affecting their attitudes towards these groups negatively (Allport et al., 1954;Esses et al., 2001). Building on previous research, we examined three factors pertaining to the individual and social self that might explain who holds polarized attitudes towards refugees and migrants: (1) education level, (2) political orientation, (3) income and perceived relative deprivation.
First, several studies linked education to refugee and migrant attitudes (Borgonovi, 2012;Easterbrook et al., 2016). Across Europe, lower education levels were found to be associated with more prejudice, xenophobia, and negative attitudes towards migration (Hainmueller & Hiscox, 2007;Hello et al., 2002). Education may be a source of social status (van Noord et al., 2019); therefore, lesseducated people may experience their position in society as less secure and may experience more competition with newcomers. We therefore hypothesized that less-educated people would be more likely to have anti-refugee and anti-migrant attitudes, whereas more highly educated people would be more likely to have pro-refugee and pro-migrant attitudes (Hypothesis 1a).
Second, political orientation may be linked to intergroup relations in society. Left-versus right-wing orientations represent an ideological division in terms of two core factors: tradition and acceptance of inequality (Jost et al., 2003;Verkuyten et al., 2016). Individuals with a right-wing orientation tend to value conservation, maintenance of the status quo, and patriotism, and to accept greater inequality.
Individuals with a left-wing orientation tend to value equality, tolerance, and egalitarianism, and to embrace socio-cultural change (Jost, 2017;Piurko et al., 2011); consequently, they are more likely to accept refugees and migrants (Davidov et al., 2008). We therefore hypothesized that people with more right-wing political views would be more likely to have anti-refugee and anti-migrant attitudes, whereas people who were more left-wing would be more likely to hold prorefugee and pro-migrant attitudes (Hypothesis 1b).
Third, earlier research found that people with higher incomes may have more positive attitudes towards refugees and migrants than people with lower incomes (e.g., Burns & Gimpel, 2000).
However, other research suggests that the subjective experience of economic hardship rather than objective measures of income is especially relevant in attitudes towards outgroups (Fetzer, 2000;Walker & Smith, 2002). The subjective experience of being more deprived than others is called relative deprivation (Guimond & Dambrun, 2002). Note, our focus here is on personal relative deprivation. People who think they are relatively deprived believe that they are worse off in society and do not benefit economically as much as others may be more inclined to experience competition for resources. Accordingly, we hypothesized that people who feel more deprived would be more likely to have anti-refugee and anti-migrant attitudes, whereas people who feel less deprived would be more likely to be pro-refugee and anti-migrant attitudes (Hypothesis 1c).

| Perceptions and experience regarding refugees and migrants
We consider people's perceptions and experience regarding refugees and migrants and the related question What are "they" like? is an important second dimension in explaining attitudinal polarization.
People tend to believe that nationality or cultural backgrounds represent a foundational fixed core that makes group members what they are (Prentice & Miller, 2007). From this perspective, national or cultural categories are viewed as real biological phenomena and are among the most essentialized categorizations (Haslam et al., 2000;Rothbart & Taylor, 1992). As a result, people tend to overestimate similarities within and differences between cultural groups (Yzerbyt et al., 2001).
When two cultures are believed to differ to a large extent, people perceive a large cultural distance (Babiker et al., 1980). The concept of cultural distance, as defined by Triandis (1994), includes several characteristics such as differences in religion, family, mother tongue, and values. For example, French host-society members perceiving more cultural distance from migrants (with respect to religion, family, etc., based on Triandis's definition), showed higher levels of prejudice toward migrants (Mahfud et al., 2015(Mahfud et al., , 2018. Similarly, among Dutch host-society members a larger perceived cultural distance (PCD) in cultural values was associated with more negative outgroup attitudes towards different refugee and migrant groups (Albada et al., 2021). When refugees and migrants are believed to F I G U R E 1 Overview the three integrated dimensions for understanding polarization in attitudes towards refugees and migrants differ considerably from the host-society, host-society members may come to believe that refugees and migrants threaten and undermine their culture (Velasco González et al., 2008). Although related, PCD differs conceptually from symbolic threat, as it is solely about perceiving differences with no particular valence attached. Symbolic threat, on the other hand, indicates the extent to which people feel threatened by differences between groups and thus conceptually represents a negative valence.
During the recent influx of refugees into Europe, many came from countries with a Muslim majority (UNHCR, 2016(UNHCR, , 2019United Nations, 2017). At the same time, islamophobia, a particular form of prejudice and racism against Muslims and Islam, has grown in Europe (Croucher & Cronn-Mills, 2011;SETA, 2017). Islamophobia and negativity towards Muslim communities tend to target "Islamic culture" (Bruneau et al., 2018). This is also the case in the Netherlands, where the majority of Dutch citizens are opposed to Muslim practices (Adelman & Verkuyten, 2020). Importantly, not all refugees and migrants are Muslim, but people nonetheless often infer such a link (e.g., Pedersen et al., 2006;Wike et al., 2016). We therefore hypothesized that people who perceive more cultural distance from Islam would be more likely to have anti-refugee and anti-migrant attitudes, whereas people who perceived less cultural distance from Islam would be more likely to be pro-refugee and pro-migrant (Hypothesis

2a).
Contact experience with refugees and migrants may offer additional insights into why host-society members may have polarized attitudes. Contact experience can enhance knowledge about outgroups and increase empathy and perspective-taking, thereby improving attitudes towards outgroups (Dahl et al., 2019). However, intergroup contact does not lead to more positive outgroup attitudes per se. Positive and meaningful contact, such as friendships, is likely to improve outgroup attitudes (Allport et al., 1954;Pettigrew & Tropp, 2008); however, it is unclear how often host-society members have such contact (e.g., Dixon et al., 2005). We thus assessed the frequency of contact that host-society members have with refugees and migrants. We hypothesized that people with little or no contact would be more likely to have anti-refugee and anti-migrant attitudes, whereas people with more frequent contact would be more likely to have pro-refugee and pro-migrant attitudes (Hypothesis 2b).

| The societal context
The intergroup relationship between host-society members and refugees and migrants does not take place in a vacuum. Societal context is therefore the third dimension, we argue, that plays a role in attitudinal polarization, as the (perceived) state of society may act as a catalyst for attitudes towards refugees and migrants (Esses et al., 2005). Here, the relevant questions are: How is society doing?
How capable is it of dealing with refugees and migrants? We focused on societal discontent as an indicator of the societal context. Societal discontent is defined as a collective global evaluation that society is not doing well, and as the belief that most people judge society negatively (Steenvoorden, 2015). Interestingly, although people in European countries report high levels of wellbeing, they may at the same time endorse pessimistic outlooks on society; thus, societal context should be considered as a separate dimension in attitudinal polarization (van der Bles et al., 2015). The evaluation that society is not doing well may spread to a negative evaluation of the government's functioning, including the belief that the government and society are incapable of properly dealing with migration. Previous research in the Netherlands has indeed linked societal discontent to concerns about migration policies (Kuppens et al., 2020;Postmes et al., 2017). We therefore hypothesized that people who experience more societal discontent would be more likely to have anti-refugee and anti-migrant attitudes, whereas people who experience less societal discontent would be more likely to be pro-refugee and pro-migrant (Hypothesis 3).

| THE CURRENT RE S E ARCH
In the current research, we first investigated whether and to what extent Dutch society is polarized in its attitudes towards refugees and migrants. Second, we investigated the links between three dimensions with polarized attitudes towards refugees and migrants.
More precisely, for the dimension "individual and social self", we investigated how education level, political orientation, income and personal relative deprivation were linked to polarized attitudes. For the dimension "perceptions and experience regarding refugees and migrants", we examined PCD from Islam and contact experience.
For the societal context dimension, finally, we researched the role of societal discontent. This two-step approach aimed to increase our understanding of who holds polarized attitudes and what psychological factors may be associated with attitudinal polarization in society. We tested our hypotheses in a survey among a large-scale representative sample of the Dutch population. reported it as being below average, 22.7% as average, 30.6% as above average, and 19% did not respond. A small proportion (6.1%) had a migrant background (at least one parent born outside the Netherlands), most of whom came from Europe (59.1%), followed by Asia (23.5%), South America (12.2%), Africa (3.5%), and North America (1.7%). About half of the participants (55.1%) were not religious. This is consistent with the 49% reported by the Dutch Central Bureau of Statistics (CBS, 2018c). Of the religious people, most categorized themselves as Roman Catholic (22.1%), followed by Protestant (11.3%), Dutch Reformed (5.7%), and other (5.7%).

| Participants and procedure
The mean political orientation lay around the mid-point of the scale (M = 4.09, SD = 1.24, range: 1-7). Both sides of the political spectrum were well represented; 30.8% were on the political right, 42.3% in between, and 26.9% on the left. Most of the participants were employed full-time (30.9%), followed by part-time (19.7%), retired (19%), unemployed (8.4%), homemakers (8.1%), students (6.5%), entrepreneurs (3.7%), and in between jobs (3.5%). Participants gave their informed consent at the start of the study and were debriefed, thanked, and financially compensated at the end.

| Measures
We classified items into scales by carefully inspecting them and selecting those that would fit theoretical definitions based on the relevant literature (see online Appendix for all items used). The scales ranged from 1 (totally disagree) to 5 (totally agree), unless otherwise indicated.
To avoid forced choices, participants were given the opportunity to respond with 'I do not know' or "not relevant"; these responses were coded as missing data. The proportion of these responses was below 10% across items, which is similar to the percentage of missing values if people are not forced to answer each item and is an acceptable proportion of missing data (DeRouvray & Couper, 2002;Newman, 2014).

| Polarization in attitudes towards refugees and migrants
To investigate whether and to what extent people were polarized in their attitudes towards refugees and migrants, we included three different aspects. First, participants indicated on one item their general negative view of migration on a scale from 1 (very positive) to 5 (very negative). Second, we assessed acceptance of refugees with five items (example item "People must be able to flee to countries, including the Netherlands, to escape from war or persecution").
Higher values indicated more acceptance towards refugees. The scale including all five items showed a high reliability (α = 0.86).
Third, we measured perceived consequences of migration with seven items, in which higher values indicated more negative perceived consequences of migration (example item "Immigration is bad for the Netherlands and divides society into opinions and beliefs that are strongly opposed to each other"). Three of the seven items were positively framed and were reverse coded. The scale showed a good reliability (α = 0.79).

| The individual and social self dimension
To capture the dimension of the individual and social self, participants indicated their educational background, political orientation, income, and perceived relative deprivation. They reported their highest completed level of education, which we coded into lower, medium, and higher education using the standards of the Dutch Central Bureau for Statistics (CBS, 2018b), which corresponds to the International Standard Classification of Education (ISCED) developed by UNESCO (see online Appendix). Participants also rated their political orientation on a scale from 1 (very left) to 7 (very right). To measure income level, participants were asked whether their gross household income per year was below, at, or above the Dutch average (with reference numbers in euros illustrating the response categories).
We measured personal relative deprivation with two items, in line with Guimond and Dambrun (2002) ("For someone like me, it is becoming more difficult to succeed in the Netherlands" and "I am benefiting personally from the growth in the Dutch economy (reversed)"). Higher values indicated more relative deprivation. The two items correlated significantly and were combined (r = 0.40, p < .01).
In social psychology, relative deprivation is often approached as a group-level variable. However, these two items were formulated on a more individual level, therefore we refer to it as personal relative deprivation.

| Perceptions and experience regarding refugees and migrants dimension
We measured perceived cultural distance (PCD) from Islam with two items ("Islam and Dutch society are not compatible with each other" and "Most Muslims have similar values to me (reversed)"). Higher values indicated greater PCD. The two items correlated and were combined (r = 0.59, p < .01). For frequency of contact, participants indicated how often they had had contact with refugees from the Middle East, Africa, and other regions, and with migrants from Turkey, Morocco, and Eastern Europe, on a scale from 1 (never), 2 (less than once a year), 3 (once or a few times a year), 4 (once or a few times a month), to 5 (every week or more often). We combined the six items into one mean score to indicate frequency of contact with refugees and migrants.

| The societal context dimension
We measured societal discontent (Steenvoorden, 2015;van der Bles et al., 2015) with two items ("In general, did Dutch society improve, stay the same or worsen over the past year?", "In general, did the Dutch economy improve, stay the same or worsen over the past year?"). The scale ranged from 1 (improved) to 2 (about the same) and 3 (worsened). The two items correlated significantly and were combined (r = 0.41, p < .01).

| Analytic strategy to investigate polarization
To examine whether and to what extent Dutch society is polarized in its attitudes towards refugees and migrants, we conducted a latent profile analysis (LPA) in R software using tidyLPA (R Development Core Team, 2010;Rosenberg et al., 2019). LPA is a type of classification analysis that tests whether relatively homogeneous groups (profiles) can be identified based on observed values (Oberski, 2016). It tests whether structural groups exist in the data, whereby individuals show coherence with some individuals while simultaneously showing discrepancy towards others. It is therefore an excellent assessment of polarization, as we and other researchers have defined it (e.g., Bauer, 2019;DiMaggio et al., 1996).
Where no structural groups could be identified, this suggests that there is no polarization. Where structural groups are identified, this suggests there is evidence of polarization in society. For the LPA, we included the attitudinal polarization variables, which were as follows: a negative view of migration, acceptance of refugees, and consequences of migration. This analysis could only be computed for participants with no missing data. Thus, the sample size for this analysis was N = 1645. Little's test suggested that the incomplete cases that could not be included due to missingness were completely at random (MCAR). Furthermore, the distributions and the characteristics of people with missingness were similar to people with no missingness (see Supporting Information for details).
In LPA, a range of models with an increasing number of profiles are tested, making it a data-driven method. First, Bayesian information criterion (BIC) and Akaike information criteria (AIC) indicate the fit of the model, with the lowest numbers indicating the best fit (Norusis, 2011). Second, the entropy statistic indicates how well the model can divide participants into different profiles without overlap in or exclusion from profiles. Low entropy indicates that one participant can be classified into more than one profile. Entropy scores of 0.80 and higher are considered appropriate. Next, the bootstrap likelihood ratio test (BLRT) tests whether models with one additional profile outperform the previous model. Finally, to further validate the assessment of polarization and investigate the extent of polarization, we conducted a multivariate analysis of variance (MANOVA) to test differences between identified profiles and their effect sizes in order to evaluate the meaning of the profiles and the extent of their disagreement (Burns & Burns, 2008;Everitt et al., 2011).

| Empirical test of polarization
The LPA showed that a model with four profiles best fitted the data; it had the lowest BIC and AIC, and a significant BLRT (see Table 1). It also had an entropy level of 1.00, which indicated that all participants included in the analysis fitted into a profile and could only be fitted into one profile. Next, to further examine the different groups and the extent of polarization, we conducted a one-way MANOVA on the LPA variables. See Table 2 for the descriptive statistics of the groups and Table 3  First, there was a group (31.8%) that we labelled critical. While individuals in this group had a generally negative view of migration, they were neutral about accepting refugees into the Netherlands and they tended to believe that migration had somewhat negative consequences for society. Although this group did not take an extremely negative stance, it did consider refugees and migrants as a critical issue that may create societal problems. The second group (33%) in Note: Inspection of the best type of model revealed that a model allowing for varying means while holding variance and covariance equal fitted the data best (method "EEE"). **p < .01.

TA B L E 1 Model fit indices of latent profiles
the middle could be described as lenient. This group had a neutral view towards migration, leaning towards acceptance of refugees.
Furthermore, this group was relatively undecided regarding the consequences of migration for society. Although this group could not be described as positive, its members seemed to display some level of tolerance towards refugees and migrants. All in all, these results confirmed that Dutch society is polarized in its attitudes towards refugees and migrants as we found clearly separated structural groups, two of which two were at opposite ends of the spectrum and two in the middle which were divided, suggesting disagreement across all four groups.

| The three-dimensional model of attitudinal polarization
Second, to understand who has polarized attitudes and what was associated with these attitudes, we tested our integrated three- To assess whether the three dimensions could uniquely help to explain polarization, we computed Nagelkerke's pseudo R 2 (R 2 N ) and −2 log-likelihood (−2LL) of each step. Pseudo R 2 N indicates the amount of variance explained by the independent variables regarding the differences between the groups and thus provide the effect size of the model. R 2 N ranges between 0 and 1 while taking the sample size into account (Nagelkerke, 1991). In addition,

TA B L E 3 Latent profile analysis validity test -MANOVA results
and explained a substantial proportion of polarization in attitudes towards refugees and migrants (R 2 N = 0.24). Furthermore, the −2LL of the model (1,217.88) was significantly lower than the −2LL of the intercept-only model (1622.22, p < .001). Education, political orientation, and personal relative deprivation were significant predictors of polarization in attitudes, whereas income was not.
People with a lower level of education, a more right-wing orientation, and who experienced more relative deprivation were more likely to have anti-refugee and anti-migrant attitudes, whereas people who had higher education levels, a more left-wing political orientation, and who experienced less relative deprivation were more likely to hold pro-refugee and pro-migrant attitudes (confirming Hypotheses 1a-c).
Step 2, testing the dimension of perceptions and experience regarding refugees and migrants, was also significant [χ 2 (6) = 727.51,  Note: PCD is an abbreviation for perceived cultural distance. Political orientation ranged from 1 (very left) to 7 (very right). Societal discontent scale ranged from 1 to 3. Other scales ranged from 1 to 5.

TA B L E 5 Means (M) and standard deviations (SD) of profile groups
should be noted that the effects for contact were small and the majority of participants had few or no contact experiences. Moreover, education was only marginally significant in this step (p = .065).
Step 3 of the analysis, testing the dimension of the societal context, was also significant [χ 2 (7) = 707.63, p < .001]. The −2LL of the model (3,136.08) was significantly lower than the −2LL of the intercept-only model (33,843.71, p < .001). As expected, societal discontent significantly added to explaining polarization in attitudes towards refugees and migrants (Hypothesis 3). People who were unhappy about Dutch society were more likely to have anti-refugee and anti-migrant attitudes, whereas people who thought that Dutch society has been improving were more likely to hold pro-refugee and pro-migrant attitudes. However, a comparison of R 2 N from Step 2 (R 2 N = 0.41) with Step 3 (R 2 N = 0.41) showed no increase, suggesting that the added effect to the overall model was limited. Education and contact were marginally significant in this step (p = .055 and 0.049, respectively).
To test the strength of association of each factor with polarization, we examined the odds ratios (OR), which tell us how much the odds that a person will move to a higher group if there is one unit increase in the independent variable increase or decrease (see Step 3, Table 6). An odds ratio > 1 indicates that the factor is associated with the odds of belonging to a higher group. An odds ratio < 1 indicates that the factor predicts that individuals are more likely to belong to a lower group. An odds ratio of 1 indicates that the factor is not associated with polarized group membership. For example, someone who scores "2" on PCD from Islam is more than half as likely (OR = 0.35) to have pro-refugee and pro-migrant attitudes than someone who scores "1". Conversely, someone who scores "1" on PCD from Islam is about twice as likely to be pro-refugee and pro-migrant than someone who scores "2".
Interestingly, the ORs and the log odds showed that PCD from Islam was a particularly important indicator of polarization. Finally, using a MANOVA for the continuous variables, we found significant differences for all factors (see Table 7). Again, the differences for PCD from Islam were especially noticeable; all groups differed from each other and the effect size of the differences was ηp 2 = 0.33, which can be considered large. In contrast, effect sizes of the differences for the other factors ranged between ηp 2 = 0.04 and ηp 2 = 0.14, which are small to moderate effects. Closer inspection of the means of the two middle groups showed that they scored fairly similarly on the factors, except for PCD from Islam (with a difference

TA B L E 7
Differences between LPA groups on the continues dimension variables of more than half a scale point). Thus, PCD from Islam was the main factor that explained why critical and lenient people differed from each other.

| D ISCUSS I ON
The goal of the current research was to investigate whether and to what extent Dutch society is polarized in its attitudes towards refugees and migrants. Second, we aimed to understand who has polarized attitudes and what psychological factors may be associated with polarization by testing an integrated three-dimensional model. Using a large representative sample of the Dutch population, our findings suggested that Dutch society is polarized in its attitudes towards refugees and migrants. We found two substantial polarized groups at opposite ends of the spectrum, with attitudes that were either clearly anti-(16.5%) or pro-(18.7%) refugees and migrants. The majority of people fell in the middle of the spectrum. However, the middle group was not a single neutral group but instead was divided into two groups, one critical (31.8%) and one lenient (33%). Moreover, the three dimensions of the integrated model (the individual and social self, perceptions and experience regarding refugees and migrants, and the societal context) all played a unique role in understanding polarized attitudes in society.
These results are discussed in more detail below.

| Theoretical and practical implications
The Interestingly, through this approach we found that besides two groups at the more extreme ends of the attitude scales, the middle was also clearly divided into two separate groups, neither of which could be clearly described as neutral. The critical group seemed to lean towards negative attitudes and the lenient group resembled the critical group more than it did the pro-refugee and pro-migrant group. This finding might have been missed if we had used a distributional approach to polarization.
The absence of one large, neutral group may be a cause for concern considering that people with stronger attitudes are less likely to change their attitudes, even in light of substantial evidence to the contrary (Holland et al., 2002;Leeper, 2014). Currently, there seems to be a trend in Western societies in which people decreasingly interact with individuals having opposing opinions (Frimer et al., 2017;McCoy et al., 2018). Other research demonstrated that brief group discussions with like-minded individuals makes people more confident in their beliefs and stereotypes of outgroups and stimulates norms to engage in discrimination (Myers & Bishop, 1970;Smith & Postmes, 2011). Altogether, this may in turn create a type of polarization that could have detrimental effects on social harmony as well as on the target groups (refugees and migrants) of the polarized debate (Lelkes, 2016). Yet, stating that society is polarized warrants some caution, as this may suggest a social norm of choosing sides, a pressure that many Dutch people already feel (SCP, 2019).
Our findings are in line with other findings from the Netherlands.
The results of a national survey conducted in the Netherlands over a two-year period (2017-2019) also showed that a majority of people were concerned about the impact of refugees and migrants (Kuppens et al., 2020;Postmes et al., 2017). Importantly, although only a minority were against offering temporary asylum to refugees and seemed to be driven by xenophobic motives, a majority also tended to be critical and concerned about the integration of refugees and migrants and the consequences that this may have on Dutch cultural identity (Kuppens et al., 2020). Research on attitudes towards Muslims and their cultural practices also found four distinct groups in Dutch society (Adelman & Verkuyten, 2020), with two groups that either generally disliked or liked Muslims and their practices at opposite ends of the spectrum.
The middle was divided into one group that was intolerant towards some Muslim practices, and another that was intolerant towards most Muslim practices, without necessarily displaying prejudice.
In addition, the current research suggests that polarization is a complex phenomenon. Although many psychological factors previously identified in the literature may explain and predict who has polarized attitudes, we have added to earlier literature by organizing these into three dimensions that all uniquely add to our understanding of polarized attitudes towards refugees and migrants in society, and which all have their own theoretical and practical implications in the context of attitudes towards migration in Western society.

| The individual and social self dimension
Regarding the dimension of the individual and social self, we found that people who were less educated, 1 more right-wing, and who experienced more relative deprivation were more likely to have anti-refugee and anti-migrant attitudes. Conversely, people who were more highly educated, more left-wing, and who experienced less relative deprivation were more likely to hold pro-refugee and pro-migrant attitudes (supporting Hypotheses 1a-c). These findings provide insights into who may have polarized attitudes. For example, since people who are more right-wing are more focused on tradition, conservation, and authoritarianism, they may see migrants and refugees as a challenge to their own position in society and, therefore, be more likely to take an anti-migration stance (Canetti et al., 2016;Esses & Jackson, 2008).
1 Note that when factors from the other dimensions were added to the model, education was only marginally significant, with a small OR.
Populist parties are especially inclined to recognize and confirm such subjective perceptions and feelings of threat, thereby potentially increasing their appeal (Major et al., 2018;Mols & Jetten, 2014).
Altogether, the results suggest that people's perceptions of themselves and their social group membership (e.g., being highly educated) are linked to polarized attitudes towards refugees and migrants. This aligns with the theoretical reasoning of the social identity approach, which argues that social group membership plays an important role in how people perceive themselves and others (Hornsey, 2008;Tajfel & Turner, 1979). Certain groups may have more disadvantaged positions in society, as a result of which they experience increased competition and threats from outgroups, which can explain their defensive reactions, such as rejecting refugees and migrants; this seems to be especially the case for individuals who identify more strongly with their group (Allport et al., 1954;Riek et al., 2006). One practical implication of the above-mentioned results is the need to affirm a safe position in society for majority members as a potentially useful way of reducing negative attitudes towards refugees and migrants (e.g., Badea et al., 2020).
Moreover, the finding that personal relative deprivation (the subjective experience of lacking prosperity) was more strongly linked to attitudinal polarization compared to income is in line with previous research showing that it is the belief or feeling of being deprived that fuels fear and uncertainty (Jetten et al., 2015). A similar process occurs for people who are relatively wealthy (relative gratification), where the fear of losing their obtained wealth is a reason to reject refugees and migrants (Jetten, 2019). Refugees and migrants may be seen as a source of such perceived deprivation, and therefore become targets of rejection (Jetten et al., 2015). Despite research suggesting that migration is beneficial for the economy of host countries and that asylum seekers do not seem to cause negative effects (d'Albis et al., 2018), many people view refugees and migrants as a challenge to society (Riek et al., 2006).

| Perceptions and experience regarding refugees and migrants dimension
The findings of the current research suggest that people who perceived a large cultural distance from Islam and who had less frequent contact with refugees and migrants were more likely to have anti-refugee and anti-migrant attitudes. Conversely, people who perceived less cultural distance from Islam and had more frequent contact were more likely to hold pro-refugee and pro- We also found an association between frequency of contact and polarized attitudes, although it was small and only marginally significant. Note, however, that in the current research, people had almost no contact with refugees and migrants. Other research recently conducted in the Netherlands has suggested potential beneficial effects of intergroup contact. More precisely, Dutch residents living close to asylum centres were generally positive (or at least not negative) towards refugees and asylum centres (Kuppens et al., 2020). Research also suggests that contact needs to be positive and meaningful (such as friendship) in order to substantially improve attitudes towards outgroups and this may not have been the type of contact people had in the current context (e.g., De Coninck et al., 2020;Pettigrew & Tropp, 2008).

| The societal context dimension
Finally, and as expected, the societal context in which the intergroup relationship between Dutch host-society members, refugees, and migrants is embedded is related to attitudes towards migrants and refugees (Hypothesis 3). More precisely, people who experienced more societal discontent were more likely to hold anti-refugee and anti-migrant attitudes, whereas people who experienced less societal discontent were more likely to be pro-refugee and pro-

| CON CLUS IONS
To conclude, Dutch society seems to be polarized with respect to attitudes towards refugees and migrants, with two substantial groups at opposing ends of the attitude spectrum (anti and pro) and two distinct groups in the middle, who are either critical or lenient towards refugees and migrants. Around half of the respondents took a more negative stance (anti-refugee and migrant and critical), a substantial group was lenient, and only one group (less than 20% of the respondents) could be identified as clearly pro-refugee and pro-migrant. These results have clear societal relevance, as refugees and migrants facing negative attitudes may also experience social discrimination. In fact, refugees and migrants are clearly disadvantaged in the labour market (Dumont et al., 2016), and suffer significantly more (mental) healthcare issues than members of the host-society (Davies et al., 2006). We hope that our integrated three-dimensional model will inspire future research to develop a better understanding of polarization in society, its complexity, and possible ways to intervene.

ACK N OWLED G EM ENTS
We thank the network Stay Human, the Netherlands, and especially Jorrit Hoekstra, for sharing the data with us, which was collected by in June 2017 by Ipsos for the report "Nationale monitor polarisatie en beeldvorming vluchtelingen een onderzoek ten behoeve van een effectievere, depolariserende strategie over vluchtelingen" (2018, project 16071564). While this data set was used for the reported analysis, the responsibility for the content as presented in this article lies solely with the authors. Moreover, we like to thank Anne Marthe van der Bles for her comments on the manuscript.

CO N FLI C T O F I NTE R E S T S
The authors declare that they had no conflicts of interests with respect to their authorship or the publication of this article.

E TH I C A L S TATEM ENT
The research for the current article was in accordance to the ethical guidelines of the Declaration of Helsinki, the National Ethics Council of Social and Behavioral Sciences, and the APA.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data used in the current research will not be made public as the authors do not have legal ownership of the data. Stay Human and the Dutch Council for Refugees/VluchtelingenWerk Nederland allowed the authors to use the data for this publication. The data will be archived by the University of Groningen for 10 years.

S U PP O RTI N G I N FO R M ATI O N
Additional supporting information may be found online in the Supporting Information section.

ITEM S
All items ranged from a scale from 1 (totally disagree) to 5 (totally agree) unless otherwise indicated.

Negative view of migration
Higher ratings indicated a more negative view on migrations. The scale ranged from 1 (very positive) to 5 (very negative).
• Do you think that immigration had a positive or negative impact on the Netherlands?

Acceptance of refugees
Higher ratings indicated more acceptance of refugees and refugee shelters in the Netherlands.
• We must close our borders completely for refugees-we cannot accept any more at the moment. (reversed) • People must be able to flee to countries, including the Netherlands, to escape from war or persecution.
• We have to accept refugees in the Netherlands because it is part of our country's culture and history to welcome people who flee war and violence.
• The Netherlands has the economic and financial resources to welcome refugees, and therefore has an obligation to do so.
• For decades, people with different backgrounds and with different beliefs have come to the Netherlands. But as long as they work hard, learn the language and contribute to society, there is room for them here.

Perceived consequences of migration
Higher values indicated more negative perceived consequences of migration.
• Immigration is good for Dutch cultural life and makes the Netherlands a livelier and more exciting place to live. (reversed) • Immigration is bad for the Netherlands and divides society into opinions and beliefs that are strongly opposed to each other.
• Refugees who come to the Netherlands these days ensure that the Netherlands is more open to new ideas and cultures. (reversed) • Immigration is good for the Dutch economy and provides new skills, new opportunities and stimulates success. (reversed) • Immigrants in the Netherlands have made it more difficult for Dutch people to find employment.
• Immigrants claim benefits and social benefits, even though they do not contribute.
• Immigrants are given priority over our own Dutch population when it comes to social benefits, housing, or use of public facilities.

Political orientation
Scale ranged from 1 (very left) to 7 (very right).
• In terms of your political beliefs, where would you classify yourself?

Income
• Nowadays people often use the term modal income. Modal means that you earn between 28,500 and 34,000 euros a year, including holiday pay. Which of the following situations is most applicable to your household income?
1. Below average (less than 28,500 euros gross per year) 2. Modal (between 28,500 and 34,000 euros gross per year) 3. Above average (more than 34,000 euros gross per year)

Personal relative deprivation
Higher ratings indicated higher experience of personal relative deprivation.
• For someone like me it is becoming more difficult to be successful in the Netherlands.
• I am benefiting personally from the growth in the Dutch economy. (reversed)