The public view of immigrant integration: multidimensional and consensual. Evidence from survey experiments in the UK and the Netherlands

ABSTRACT Despite the growing interest in immigrant integration in a wide range of scholarly literature, there is less interest in how integration might be understood by the public. Using a survey-embedded conjoint experiment in the Netherlands and the UK, we ask the public what they think constitutes successful immigrant integration. We show that the public has a multidimensional view of integration, which goes beyond a simple preference for cultural assimilation. We discover that there is a remarkably stable hierarchy of preference of integration outcomes, which is a matter of wide spread consensus in both our countries among different social groups and people with different attitudes on immigration. Using the British data we also show an integration penalty for immigrants of non-white origins. Our article places public opinion of immigrant integration at the heart of a rapidly expanding research agenda into the social and political impacts of immigration.


Introduction
In the face of growing and diversifying migration flows, immigrant integration is increasingly present in academic, policy and public debates and this trend is likely to continue in the foreseeable future (Dancygier and Laitin 2014;Alba and Foner 2015). These debates are especially pronounced as scholars seek to map and explain if and why multiculturalism has fallen out of favour as a model of integration. The rise of 'home-grown' terrorists has been taken as a signal of this apparent failure of multiculturalism and the lack of integration among Europe's growing immigrant Muslim communities has been inextricably linked with issues of terrorism in both policy and media (Bleich 2009;HM Government 2011, Sobolewska andAli 2015). Public opinion research on integration clearly supports the mainstream narrative of multiculturalism's failures, with most of it showing an almost universal preference for cultural assimilation of immigrants and disapproval of multiculturalism (Van Oudenhoven et al. 1998;Zick et al. 2001;Arends-Tóth and Van De Vijver 2003;Breugelmans, Seger, and Van De Vijver 2004). However, given the complexity that may constitute multicultural or assimilationist policy in different countries and the fact that academic studies employ many and multifaceted definitions and measures of integration, the existing public opinion research on this issue seems relatively unsophisticated. As a result it tells us very little about how the public really envisages how these processes take place, and what exactly it has in mind when it expresses its preferences for one model of integration over another.
Given this substantial gap in our knowledge, in this paper we propose to come back to basics and explore the fundamentals of how the public envisages integration without the pre-packaged and politically loaded labels of integration policy models. As a result, we ask more elementary questions. Do people use a broad range of criteria of integration used by the academic research? Or do they think exclusively of cultural assimilation in keeping with the policy trends and media accounts? Do stereotypes about Muslims, often found in the media and elite narratives, find their way into public consciousness and impact the way the public judges Muslim immigrants' integration? Finally, do they all envisage integration in the same terms, or differ according to their socio-economic profile or ideological background?
To answer these questions we use survey-embedded conjoint experiments in two countriesthe Netherlands and the UK. These two countries have been, until the mid-1980s and early 1990s, the very paragons of European multiculturalism, but have more recently experienced a significant policy change towards a more assimilationist model. The two countries also share a large immigrant Muslim community, which is an important factor in these countries' departures from multiculturalism. We deliver three important findings, all of which open new lines of inquiry into integration. First, we show that there is a clear hierarchy of preference for integration outcomes, which places cultural and social indicators of integration as the most important universally, but also shows a complexity that reflects the multidimensionality and idiosyncrasy of the process as we see it described in some academic studies (Bean et al. 2012;Lessard-Phillips 2015). Interestingly, the key aspect of Muslims' contested place in European societies promoted by the media and politicianstheir religion (Betz and Meret 2009)held only a little sway over how the public perceived Muslims' integration; although another construed source of Muslim's 'otherness' their attitudes towards women (Bracke 2012)mattered a lot more. Second, we discover that this hierarchy of indicators is remarkably stable across the usual demographic and attitudinal differences between people. This consensus extends across countries as we find nearly identical results in both our case studies. Third, in the UK, we show that migrants' origin matters to how well integrated they are perceived to be. Although the importance of different integration dimensions remains unchanged across origins, Muslim and African-origin immigrants are seen as less integrated than otherwise identical white immigrants. However, this prejudice is not a matter of consensus: not all social groups share it.

The multidimensionality of integration
Dualism between assimilation and multiculturalism as models of integration dominates the scholarly literature, but it is unclear whether they reflect how integration is understood by the populations of the host countries. Yet, attitudes towards assimilation and multiculturalism are almost all we know about public opinion in this area. The existing literature on public perceptions of integration focuses on the relative popularity of these two models, usually offered as broadly sketched and pre-packaged alternatives with no possibility for overlap or interaction between the different dimensions of integration. From this research it emerges that the host populations generally express a preference for cultural assimilation over multiculturalism (Van Oudenhoven et al. 1998;Zick et al. 2001;Arends-Tóth and Van De Vijver 2003;Breugelmans, Seger, and Van De Vijver 2004). There also seems to be a linear understanding of assimilation in the eyes of the public: losing one's own ethnic culture and values is perceived as inevitable as one adopts those of the majority culture (Arends-Tóth and Van De Vijver 2003). Looking at studies conducted in various national contexts, the public perceives multicultural policy as a gateway to separate ethnic rather than common national loyalty, and these are mostly seen as mutually exclusive (Ho 1990;Zick et al. 2001;Arends-Tóth and Van De Vijver 2003;Breugelmans, Seger, and Van De Vijver 2004). However, this linear and inevitable pathway towards assimilation may be an artefact: these studies usually present pre-defined options, rather than multi-dimensional and 'messy' combinations to choose from.
The scholarly literature on integration has initially employed a similarly one-directional understanding of integration, in which distinct areas played different roles in the inevitable, linear process: with linguistic assimilation as the first necessary, but insufficient condition of integration in other domains, and a structural integration as the keystone and a turning point for achieving full assimilation (Gordon 1964). Although this linear progress is significantly impeded by various disadvantages including prejudice, success in one domain is expected to lead to success in other domains (Gordon 1964;Park and Burgess 1969). With time this perspective fell out of favour (although see new assimilation theory for a defence Alba and Nee 1997;Alba and Nee 2003;Waters et al. 2010).
Thus the later theories of integration, such as the segmented assimilation theory, argue that upward socio-economic integration into the white middle-class group may not necessarily be followed by cultural integration, with some minorities maintaining close links with their ethnic, cultural or religious communities and identities (Portes and Zhou 1993;Kymlicka 1995;Bloemraad, Korteweg, and Yurdakul 2008;Fokkema and de Haas 2015). More recent research has gone even further by arguing that in fact there may be a trade-off between some dimensions of integration. Gans (1992) argues that delayed acculturation may benefit the children of immigrants from disadvantaged backgrounds as acculturation into the mainstream US culture raises these children's expectations beyond what they can achieve in the discriminatory labour market, making them unable to gain employment at all. A parallel example can be found in other dimensions: Maxwell (2012) has found that ethnic minorities that are better socio-culturally integrated are less successful in the political and economic realms. Others have argued that spatial segregation of ethnic minorities in specific urban areas, which is usually presented as lack of social integration, could foster economic (Portes 1987;Balakrishnan and Hou 1999) and political integration (Leighley 2001) or coexist with positive results in these and other arenas (Musterd 2003;Phillips 2007). This idiosyncrasy of integration creates a multidimensional picture in which different routes result in differentially defined integration success on multiple and non-linear axes (Bean et al. 2012;Lessard-Phillips 2015).
Could these complex conceptualisations of this increasingly important social process vary so significantly between the academic and general publics? We would argue that such disjoint between the complex real-life issue and public opinion is unlikely, and so we posit that the one-dimensional account of preference for cultural assimilation is a likely result of unsophisticated measurement. We therefore propose that, using a method that allows the public to simultaneously consider many indicators without forcing a presumed relationship between them and thus to express a more multifaceted view of integration, we will see some degree of complexity.
H1. The multidimensionality hypothesis: Indicators of cultural, social, economic and political integration will be all considered in defining integration success in a fashion that allows for a mix and match approach reflecting uneven integration along these dimensions.

Divided public opinion
The literature on attitudes towards immigration identifies that populations are divided in their perceptions of ideal levels of immigration as well as on whether immigration is a good thing. It is therefore likely that the views of integration will be divided along similar lines. Existing research proposes a few explanatory hypotheses about how these attitudes are distributed and how they are associated with particular demographic differences, especially levels of education, but also socio-economic status (SES). These attitudinal tendencies can be broadly grouped into two categories that reflect different sources of perceptions of threat of immigration: economic and cultural ethnocentrism.
Economic ethnocentrism may spring from a sense of economic threat (Dancygier and Donnelly 2013), which broadly overlaps with the realistic threat theories (Stephan and Stephan 2000) and predicts that economic worries will result in in-group preference for access to scarce resources such as jobs or welfare. Previous research found that the economic threat arguments (Scheve and Slaughter 2001) applied differently for different socioeconomic groups. While individuals with lower SES were worried about employment competition with immigrants, individuals with higher SES were concerned about their tax burden and showed a preference for immigrants paying more taxes and receiving fewer benefits (Hanson, Scheve, and Slaughter 2007). However, some argue that a more generalised, or sociotropic, economic attitude matters more for immigration opposition (Hainmueller and Hopkins 2015). Since the subject of this concern is still economic, we will include this in the economic explanations, but dub it 'economic ethnocentrism': a generalised economic preference for natives.
The existing research on attitudes towards multiculturalism finds that the economicbased theories do not contribute much to the public opinion on integration (Ho 1990), but we believe this is because most of this research presented a multiculturalist model for public evaluationoften in direct contrast with assimilationand therefore did not explicitly tackle the socio-economic attainment dimensions of integration. In line with research on immigration attitudes, which puts particular weight on these dimensions, we will therefore expect that both the individuals' generalised economic attitudes and economic position will explain their preference for socio-economic dimensions of integration.
H2: Preference for socio-economic dimensions of integration will correlate with economic ethnocentrism of the individual respondent.
H2a: More importance being placed on the tax/benefits contribution dimension of integration will correlate with higher SES of the respondent.
We dub the second form of ethnocentrism 'cultural ethnocentrism' (Kinder and Kam 2010), although we do understand it to include related notions of symbolic threats (Stephan and Stephan 2000) such as national belonging and identity, culture and values. According to these theories, individuals perceiving high level of symbolic threat have negative attitudes towards members of the out-group (McLaren and Johnson 2007). Unsurprisingly, research on attitudes towards integration found that ethnocentric beliefs in particular explain the preference for cultural assimilation as a way of achieving integration and negative attitudes towards multiculturalism (Ho 1990;Zick et al. 2001). However, since the support for cultural assimilation and the other dimensions of integration have not actually been compared, we can add to this body of literature by hypothesising that cultural ethnocentrism will explain the preference for cultural, but not the other, dimensions. On the other side of the coin, those with lower levels of cultural ethnocentrism will be less insistent on cultural assimilation and perhaps even less demanding of immigrants altogether. Since we expect those who are more educated to be less culturally ethnocentric (Coenders and Scheepers 2003), we will also find this pattern for those with university degree. H3: Preference for cultural, but not other, dimensions of integration will correlate with cultural ethnocentrism attitudes of respondents.
H3a. Those with university degree will be less 'demanding' of immigrants, particularly in regard to cultural integration.

Integration in comparative perspective: the UK and the Netherlands
The selection of the countries is dictated both by their similarities and their differences that can shed light on the generalisability of attitudes towards integration. Both the Netherlands and the UK have a comparable experience with the post-war waves of post-colonial and labour immigration and a current influx of EU immigrants from Eastern Europe (Lessard- Phillips, Fleischmann, and Van Elsas 2014). Similarly also, the immigrant minorities that spark the greatest public controversy and are usually discussed in the context of integration are Muslim, although of different national origins (Adida, Laitin, and Valfort 2016). The two countries have also shared a history of a moderate form of multiculturalism (Tolley 2011), as ranked by various policy indices, including the Migrant Integration Policy Index (MIPEX) (Huddleston et al. 2011). Both countries' multiculturalism models focused on equality and anti-discrimination, particularly in employment (Joppke 2007;Statham et al. 2005;Duyvendak, Pels, and Rijkschroeff 2009). These similarities enabled us to field identical experiments in the two countries, with the only difference being the different names and national origins of immigrants used in our vignettes (see the Methods section). The use of different origins can be seen as 'essentialising' Muslims on the basis of their religion over different cultural heritage. Sadly, however, it reflects a reality in which Muslims are perceived as a homogenous group in European host societies and are stereotyped on the basis of their religion by the public narratives in both countries (Field 2007;Saeed 2007). The use of two different groups can also be defended on the basis of comparability: the numerical and politically salient position in their respective societies creates more equivalency of measurement than using the same national-origin group across the two countries.
The recent divergences between the two countries' policies offer a good test of how universal the attitudes towards integration are. Even though both countries have experienced public debate around the supposed failures of their multicultural policies and moved towards more 'assimilationist' policies especially around cultural diversity, citizenship and family reunion (Joppke 2007;Duyvendak, Pels, and Rijkschroeff 2009;Goodman 2010;Sawyer and Wray 2014), the results have been quite different. According to the Multiculturalism Policy Index (MPI) for Immigrant Minorities (Tolley 2011), in 2010 the UK still ranked as a mid-table multiculturalist regime, while the Netherlands returned to a 'weak' multicultural context, hence experiencing a more pronounced departure from this approach to integration. Similarly, in the UK the one viable far-right political party remains at the political margins while their Dutch counterparts are present in parliament. Admittedly, the UKIP's electoral popularity has been rising since 2010 and arguably proved to be a decisive factor in Britain's vote to leave the European Union in 2016. However, the fact that it did not translate into parliamentary success (because of the UK's voting system) remains a significant difference between the two countries. Since Guimond, de la Sablonnière, and Nugier (2014) show that attitudes towards multiculturalism and assimilation are strongly influenced by the current political and policy regime in the national context, we expect these differences to yield a good test of how generalisable the attitudes towards integration really are.
The role of religiosity is an important difference between the two countries. In the Netherlands, multiculturalism was explicitly organised around denominationally defined institutions (Statham et al. 2005); in the UK, religion has been long perceived as the source of political violence (first in Northern Ireland and now Islamic). Thus, religion in the UK is a politically contentious issue. In political discourses, religiosity creates a distinction between 'good' and 'bad' Muslims, with the 'good' Muslims defined as moderate: 'a moderate Muslim [keeps] one's faith private which insinuates that any Muslim who allows their Islamic beliefs to influence their political activity is extreme' (Moosavi 2015, 6).

Data and methods
Our data come from online surveys conducted in the Netherlands, by CentERdata, and the UK, by YouGov. Both surveys were fielded online but pollsters used a different methodology (see online appendix). Data collection in the two countries varied in time due to funding constraints: our Dutch data was collected in 2013, and was replicated in the UK in 2014. Our sample is 1101 respondents in the Netherlands and 1894 in the UK. The surveys included the same attitudinal controls such as general anti-immigrant sentiment and perception of cultural and economic ethnocentrism, and the usual demographic controls such as education, SES, age and gender. Weights, only available in the UK data, were used (see online appendix). A look at the two samples from the UK and Netherlands (see Table 1) reveals a number of differences between the two countries in some relevant controls. Firstly, our Dutch respondents were more likely to be men, have a degree and were on average older. Secondly, the Dutch sample was a lot more liberal in their attitudes towards immigrants. This has to be taken into account when interpreting the results and comparing the two countries.
We used an experimental approach to measure respondents' views on integration. This design has in the past added nuance to the picture of public opinion on immigration, allowed researchers to make causal inferences and is becoming increasingly popular, especially in online formats (Sniderman, Hagendoorn, and Prior 2004;Hainmueller and Hiscox 2010;Harell et al. 2012;Iyengar et al. 2013). The vignette approach is particularly appropriate to measure a potentially sensitive opinion as it is thought to be relatively unobtrusivesee Mutz (2011). Our chosen experiment is also uniquely well suited to measuring a complex issue with many potential dimensions and interactions. Following the work of Hainmueller, Hopkins, and Yamamoto (2014), we use the conjoint experiment method, based on (1) a choice and direct comparison between pairs of immigrant profiles and (2) a rating of the level of integration of various profiles. This allows identifying the relative impact of the immigrants' attributes in the judgement that the public makes about their levels of integration. In our first experiment, we present each respondent with six vignettes of an immigrant, in pairs of two, and ask them to choose the one who they think is more integrated. As a robustness check, in our second experiment, we show each respondent separately three vignettes of immigrants, which we ask them to rank on a 0-10 scale on how well integrated they are (with 10 being the most integrated). We have departed from the original method as described by Hainmueller and colleagues (2014), who presented attributes of immigrants in tables, by embedding our varying attributes within a traditional vignette look, thus losing the ability to randomly vary attribute order, but we feel making our measurement more unobtrusive and more in line with the classic vignette treatments.
In the first set of profiles presented to respondents, all our immigrants were Muslimorigin men. They were from Morocco in the Netherlands and from Pakistan in the UK the two main groups perceived as the so-called 'problem' minority in each country (Field 2007;Sniderman and Hagendoorn 2007). Selecting Muslims only for the first treatment allowed us to test the importance of religion-based stereotypes and cultural tropes such as Muslims' alleged illiberal attitudes towards womenthat are often perpetuated in the media and elite narratives in both the UK and the Netherlands (Saeed 2007;Sniderman and Hagendoorn 2007) in a way that would be difficult if we also varied immigrant origin (as for other immigrant origins these are not salient stereotypes). All our immigrants were in full-time employment and law-obeying: we felt this would be the 'common-sense' minimum requirement for a well-integrated immigrant. We also described them all as having immigrated as children, to make sure all our immigrants would be eligible for citizenship (both the Netherlands and the UK have a five-year residence requirement) and they would be expected to have a basic knowledge of the language also traditionally perceived as a minimum requirement of integration and often also immigrant admission (Hainmueller and Hopkins 2015). Then we added nine attributes measuring main domains of integration as usually described by the literature (for a list of commonly used indicators in research on integration outcomes, see Bean et al. (2012); Alba and Foner (2015); Fokkema and de Haas (2015); OECD/EU (2015)): these include citizenship and voting habits as indicators for political integration; intermarriage and cross-ethnic friendship as social integration outcomes; high-and low-skilled occupation and tax status as measures of economic integration; language spoken at home as a measure of cultural integration. To this last category we also added attitude towards women's employment and finally level of religiosity. The lower level of participation of Muslim women in the labour market, despite increases in the level of educational attainment (Dale, Lindley, and Dex 2006;Fischer and Pastore 2015), has been highlighted (either justifiably or not) as an area of concern in terms of integration (Alba and Foner 2015), and given that the often suggested explanations for this pattern are Muslim women' traditional gender roles being incompatible with employmentthis is a relevant measure of cultural difference between Muslim immigrants and Western cultural norms.
To test the role of religion, which is usually implicated in the negative portrayals of Muslim in the West (Betz and Meret 2009), we also include the indicator of religiosity. Given the status of the last two measures as particularly stereotypical in anti-Muslim narratives, the test of whether these are present in the public mindand whether they impact on how well integrated Muslim immigrants are thought to beis particularly interesting. We varied all attributes randomly in each vignette and offered two variants for each attribute. For high-versus low-skill occupation our immigrants were builders or working in IT. Builders constitute ideal lower skilled, manual, occupation job that could nevertheless be fairly well paid (which allowed us to vary the tax contribution). IT was selected as an example of a high-skilled occupation, which nonetheless could be believable as a career including some entry-level jobs where some tax rebates/tax credits could be possible, especially as all our immigrants had children. Our immigrants also either paid a higher rate of taxes, or were receiving tax rebates/tax working credits. Our respondents were either married to a co-ethnic (as suggested by name) or a Dutch/British woman (again, as suggested by name: Dutch in the Netherlands or English in the UK). Most of our immigrants' friends were either of host or of immigrant origin. Since we assumed that some level of language competency would be a common sense minimum requirement, we varied whether the origin country or destination country language was spoken at home to the immigrants' children. Our immigrants were either of the opinion that women should stay at home or work full-time. Finally, while all our immigrants were implicitly Muslim by origin, we varied whether they were religious or not. The pairings of immigrants were randomly generated before fieldwork so that identical pairings could be excluded (see online appendix).
In the UK survey we were also able to run an additional, shorter experiment, where we varied the origin of the immigrants to see how well our findings generalise to other groups, since origin strongly influences how welcome the immigrants are. To test the robustness of our findings we designed this question in a different format and with different attributes reflecting dimensions of integration. The experiment presented each respondent with three randomly generated, non-identical, vignettes of female immigrants and asked them to rank them on a scale of 1-10 on how well each was integrated, with 10 indicating good levels of integration. Since we included different origins, for many of which religion is not salient, we excluded religiosity as an attribute; also because our immigrants were female we removed gender attitudes. We also removed voting as some of these immigrants were not eligible to vote at all unless they were citizens (which we varied). We also dropped intermarriage, which was not significant in the first experiment, and replaced our two socio-economic attributes, which also did not work, with another indicator of immigrant's status: route of migration. Consequently, we varied randomly six attributes: origin, reason for immigration, social contact in neighbourhoods and a place for socialising, citizenship and cultural integration through cooking British food. Since we had fewer attributes in this experiment, we were able to test the role that immigrant's background plays in their perceived integration. For national origin, our immigrants were either from Poland (for an unpopular white economic immigrant from Eastern Europe, following Ford 2011), Ghana (for a black immigrant, from a country not considered problematic), Bangladesh (for a Muslim immigrant), France (for a popular white immigrant) and China or Hong Kong (for a culturally distant immigrant, but not problematised in British anti-immigrant narratives, again following Ford 2011).
For reason for immigration, which is equally salient, we varied three options: to improve their living standards; to get an education (the highest status, and popular, reason for migration among the host British population); or to join family in the UK (the low-status and least popular route of migration) (Lessard-Phillips, Fleischmann, and Van Elsas 2014). To indicate civic integration we used citizenship again, however, we reframed this as an active (applying for) rather than passive (having) status. Thus we had two variants: the immigrant recently applied for citizenship, or had no plans to apply. We used two measures of social contact. Firstly, we changed the wording of this attribute from having English friends in the first experiment to living in a locality with English or immigrant neighbours in order to add a spatial dimension of integration. Secondly, to further test this influential aspect we had our immigrant socialise either in the neutral local community centre, with fellow immigrants, or in a pub (which is perceived as a mainstay of English culture). Finally, for cultural integration we have abandoned language, to test the importance of cultural integration outside of this dominant attribute. We varied the immigrant's efforts to maintain her own culture or learn about the English one, and to make it more tangible we added as evidence that they either cooked their own food, or tried to cook English dishes. We felt that giving an example of the expressed commitment was important, but we wanted to keep it fairly trivial to see how much the expression of effort alone mattered (speaking English or Dutch at home in experiment one was non-trivial by comparison). Because of the wide variety of immigrant origins in this experiment, we were unable to test the impact of any specific anti-immigrant stereotypes in the second experiment.
In our analysis, performed according to the methodology proposed by Hainmueller, Hopkins, and Yamamoto (2014), we reshape the data to increase statistical power. Since this produces repeated responses across respondents (six and three lines of data per respondent in the two experiments, respectively) we account in our analysis for the possibility that the results may be correlated across respondents by clustering the analyses by respondent. We then estimate the average marginal components effects (AMCEs) of the different attributes in a given vignette via a linear regression model. The dependent variable is the profile choice (0 = not chosen, 1 = chosen) for the first experiment and the profile ranking (0-10 rating) for the second, UK-only experiment. The independent variables are the attributes in the shown immigrant profiles in a dummy variable format, where the reference category is the attribute showing 'lower' levels of integration. The AMCEs produced with this method represent the 'average change in the probability that a profile will win support when it includes the listed attribute values instead of the baseline' (Hainmueller, Hopkins, and Yamamoto 2014, 19) and show which attributes of the profile lead respondents to choose who the most integrated immigrant is, or rank more highly as better integrated in the second experiment. This method works under specific assumptions, such as that of (1) the stability of the outcomes and the lack of carryover effect; (2) the lack of profile order effect; (3) the randomisation of attributes within profiles; (4) the completely independent randomization and (5) external validity of the profiles. These assumptions are tested in our analyses and presented in the online appendix. Even though some of the assumptions have been narrowly missed, as we show in detail, these effects were insubstantial. The results also allow for looking at differences across attributes of the profiles (via interaction effects) or across attributes of the respondents (by analysing specific sub-groups), which we will report on in the Results section.

Multidimensionality
Testing our first hypothesis, on the multidimensionality in the public understanding of integration we see that evidence presented in Figure 1 from both countries supports our prediction. Figure 1 presents the AMCEs, indicating how likely a given variant of each of the attributes was to be chosen to indicate integration success in comparison to its opposite (for example how likely an immigrant with a Dutch/English wife was to be chosen as the well-integrated one over an immigrant with a Muslim wife, all other attributes being held constant). Firstly, in both countries, the overall picture is far from a onedimensionaland uni-directionalview of integration. Although cultural dimension of integration is preferred with two dominant indicators, this is not to the exclusion of indicators representing other dimensions of integration. Two of the other strongest predictors of being perceived as integrated represent two additional dimensions of integration: political and social. Additionally, not all cultural indicators matter in the public's view. The strongest predictors of being seen as well integrated in both countries are the classic cultural indicators: speaking the host nation's language to children at home and, the often stereotyped, attitude towards women's employment. However, religiosity, the third measure of cultural integration and one we expected to be salient given the unpopularity of Islam, was not perceived as a crucial factor. Only in Britain immigrants were rewarded slightly for not being religious, just as we expected. However, the effect of religiosity was trumped by other social, cultural and political indicators of integration. Clearly the trope of Muslim religiosity as incompatible with integration (De Leeuw and Wichelen 2012) was not shared by the public.
Following cultural integration closely was social integration: having friends among natives mattered a lot, however surprisinglygiven how often it is used in academic studies as a measure of social integrationintermarriage mattered little, and only in Britain, to the public's perception of integration. The next most important dimension of integration was civic integration. Surprisingly voting, rather than citizenshipanother popular indicator used by academicshad a bigger impact on how well integrated an immigrant was thought to be.
Similarly, within the economic dimension of integration occupation, not all indicators were significant. Although whether immigrant paid taxes or received benefits was a significant predictor of them being thought to be integrated, given the huge debate over the economic contribution of immigrants and the worries over their benefit entitlements, the substantive size of the coefficient was surprisingly small. In both countries builders and IT workers were thought to be equally well integrated. Although this is surprising, given the white-collar versus blue-collar choice of occupations, we acknowledge that this must be taken with a pinch of salt: both occupations we chose specifically because they can both range from quite an advanced, highly skilled and well-paid jobs to less so. As a result, economic dimension proved to be the least important to the public.
This picture shows that the public considered more than the cultural dimension of integration, usually associated with assimilation, but also that it was ready to accept that the immigrant was integrated despite not delivering on all indicators, even within the preferred cultural dimension. These idiosyncratic perceptions of integration indicate that multidimensionality is a part of the public understanding of the processadding nuance to the simple and strongly polarised understandings shown by previous public opinion research on this (Zick et al. 2001;Arends-Tóth and Van De Vijver 2003;Breugelmans, Seger, and Van De Vijver 2004). Given how different our samples were in terms of the distribution of attitudes towards immigrants, this finding also adds an element of surprise to how little difference in attitudes towards integration we found in the two contexts, giving the first indication of consensus in the public's understanding of integration.

Divided public opinion
Now turning to how the attitudinal and demographic differences between people may impact on their vision of integration, we first examine the effect of economic ethnocentrism. In keeping with the existing literature we predicted that those who reported high levels of economic ethnocentrism will prioritise socio-economic dimensions over cultural dimensions of integration. As a measure of economic ethnocentrism we used a question 'If there are not enough jobs, employers should employ Dutch/British workers ahead of immigrants'. Responses were grouped into three categories: whether respondents totally disagreed/disagreed with that statement (low ethnocentrism); whether respondents totally agreed/agreed with the statement (high ethnocentrism); and whether respondents neither agreed or disagreed with the statement or did not know/did not provide an answer (neutral ethnocentrism). Our expectation was not fulfilled as we show in Table 2. Instead, in Britain, having English friends, speaking English at home and holding liberal attitudes towards women's employmentthe cultural and social aspects of integrationwere preferred among those expressing economic ethnocentrism. In the Netherlands, on the other hand, the only significant effect of economic ethnocentrism was on voting: those who scored low put more weight on voting.
Another relevant ideological difference is cultural ethnocentrism, which we measured with a question 'Immigration is a threat to British/Dutch culture'. We hypothesised that it will explain high levels of support for cultural dimensions of integration. Since we saw before that all members of the public have a preference for cultural dimensions of integration, we already know that this preference is largely consensual, but our analysis presented in Table 2 shows that in Britain those who were culturally ethnocentric did put more weight on the indicators of cultural integration. Also, in the Netherlands, those who scored low on cultural ethnocentrism put more weight on voting. In fact, the picture closely resembled that of the effects of economic ethnocentrism, thus making these two types of ethnocentrism functionally indistinguishable. To check whether the effect of these two attitudes is additive with each other and other anti-immigrant sentiments, we looked at the summary index of attitudes towards immigration (constructed out of the four items in the lower part of Table 1 and inclusive of the two ethnocentrism measures) and we saw that this was not the case. The only effect of adding the two types of threat together was to bring from the cusp of statistical significance two effects in the UK: those who perceived threat cared more about immigrants' tax contribution and religiosity more than those who did not perceive immigrant threat. Again however, these differences did not change the overall hierarchy of preference for integration dimensions, as this remained stable for people who did and those who did not like immigrants to start with.
Considering again a variant of the economic threat argument, one suggesting that better-off people will be more concerned with tax/benefit balance of immigrants' contributions as they themselves pay-in more; we present the breakdown of the importance of immigrant attributes by the socio-economic position of the respondents in Table 3. This expectation was also not fulfilled, as there were no significant differences between what mattered for those of higher and lower SES respondents in the Netherlands, and in Britain the only significant exception was voting, which was more important for high SES respondents in the UK.
The picture was equally flat when we looked at any possible differences between highly educated and those less educated respondents: education mattered little for people's perception of integration. Again voting mattered a little more to highly educated Dutch respondents and attitude towards women was a little more important to highly educated British respondents, but the overall hierarchy was unaffected. This strengthens the message of consensus even further, as education is a strong predictor for tolerance and more pro-immigrant attitudes.
While we cannot say that pre-existing attitudes or the social profile of respondents did not matter, their impact was small and inconsistent with existing theoretical expectations. Against our expectations, we found that both economic and cultural ethnocentrism slightly increased the importance of cultural dimension of integration. None of the attitudinal or demographic measures used changed the hierarchy of preferred indicators of integration. There is a huge public consensus over what integration involves, and it is hierarchical with cultural and social dimensions trumping the socio-economic attainment among all social groups and across all pre-existing views of immigration.

Testing robustness and generalising from Muslims to other immigrants
The first experiment focused on Muslim immigrants given their status as the primary 'problem' minorities as characterised by Western media and anti-immigrant parties (Betz and Meret 2009). However, given that we found a far-reaching public consensus on integration we wanted to see whether this consensus extended over to non-Muslim immigrants. Given that Muslims' culture and religion is stereotyped as incompatible with Western values (Betz and Meret 2009), perhaps the already anti-Muslim public would demand that they show more effort in meeting the cultural criteria of integration. For other immigrants the economic dimensions might matter more or just as much as the cultural dimension. We ran an additional experiment in the UK to test this proposition. We also wanted to use this additional experiment to see whether the results of the first experiment were not a function of the question design and wording. The results showed in Figure 2 indicate that the origin of the immigrant, as we suspected, was not inconsequential to the perception that this immigrant was integrated. While the evaluation of the level of integration of the French and Chinese immigrants were not significantly different, the three other groups all received a small 'penalty', with the Muslim immigrant being perceived as the least well integrated (all other things being equal). However, effects of origin were substantively smaller than most other treatments in this experiment. Exploring English culture and cooking English food was the most influential attribute for judging integration favourably, despite the seemingly trivial nature of the example given. However, in this experiment, it was not more important than social integration: living among English people was not statistically different from culture. Immigrants socialising in a pub were deemed better integrated than those who socialised with fellow immigrants (socialising in a neutral local community centre was also positive, but less so than in a puba stereotypical mainstay of British social life). The one difference between the two experiments run in the UK was in the value respondents attached to British citizenship. As we said earlier, the second UK experiment made having citizenship more explicitly a signal of commitment through mentioning the application process and thus making the conscious choice to obtain it more salient in our respondents' minds. We worriedit seems rightlythat since the immigrants in our first experiment arrived in childhood, their citizenship status could have been either automatic or obtained by their parents, it was not highlighting the active choice and effort of seeking citizenship. In the second experiment, in which citizenship was a matter of choice, it influenced positively the public perception of integration. What is worth underlying again is that despite a different format and wording as well as selection of attributes, the hierarchy of integration preferences and the evidence of the importance of multiple dimensions of integration came through in both of our experiments.
Given that we already saw that origin mattered, we wanted to see if it will change the hierarchy of preference for integration dimensions. Thus, we asked whether immigrants who were perceived as less 'problematic', such as Chinese, were judged on different criteria than the 'contentious' Muslim immigrants. We therefore conducted additional analysis where we tested the interaction effects between different origins (with the favoured French as a reference group) and other attributes. What we found confirmed the hierarchy of dimensions we found in the first experiment. There were no statistically significant interaction effects for neither origin (see Figure 3), indicating that the public is consistenteven if they are not even handedin how they construe integration for all immigrant groups.
Finally, looking again at the perceptions of integration among different social groups and people with different pre-existing attitudes towards immigration, we confirmed public consensus around integration: the hierarchy of the different dimensions of integration did not change whatever the breakdown of our sample along attitudinal or socio-demographic lines (shown in Table 4). However, similarly to what we found in the first experiment, the economic and cultural ethnocentrism of respondents slightly increased the importance they attached to cultural dimensions of integration. We also found that immigrants' origin as a factor in their integration success was greater among those respondents who expressed ethnocentric attitudes: they found the Ghanaian immigrant particularly less well integrated than the otherwise identical French immigrant. Social attributes of the respondents were not significant for how they perceived integration.

Conclusions
Our results can be effectively summarised as delivering three important findings. First, public opinion on integration is multi-dimensional. While we see that culture is the most common deciding factor in determining how well integrated the immigrant is thought to be, not all cultural indicators were equally valued. In particular we show that the usually stereotyped aspect of Muslim integrationtheir religiositywas comparatively unimportant for the public's view of integration. Social integration and civic integration are also valued by the public in both the UK and the Netherlands, but again not all of the indicators of them were equally influential. Therefore, our research suggests that integration is perceived by the host populations as a complex and multidimensional process, in keeping with how it is empirically observed to be from the point of view of immigrant outcomes. The fact that the public rewards those immigrants who take advantage of civic rights may also be an important factor to consider in policy circles where there is a tendency to make citizenship a reward forand not the means tointegration and thus steadily increasing the difficulty of obtaining it. Clearly the public views citizenship as a means to successful integration. Making it easier for immigrants to vote is rarely discussed as a means to achieving integration, but in the light of our findings should be considered.
Secondly, and despite the nuanced picture that goes beyond the simple preference for cultural assimilation, our results point to a very broad public consensus over what integration means. The UK and the Netherlands do not differ hugely in how immigrants are asked to integrate, despite the divergent policy contextthat is, a much more evident weakening of multicultural policies in the Netherlands compared to the UK, particularly in the past decade. Also, within these countries the usual social divisions, such as education, class and existing attitudes towards immigration, matter very little to how integration is understood. The finding that the public perception of integration is a matter of consensus among people with different socio-economic backgrounds and with varying pre-existing attitudes, as well as in two increasingly divergent national contexts, offers new opportunities for research into why integration regimes vary despite the strong public consensus.
Finally, although our initial experiment measured attitudes towards Muslim-origin immigrants, we sought to replicate these findings for immigrants of different origins. While the picture of the hierarchy of dimensions remained stable, we can see that immigrants from traditionally 'preferred' backgrounds (Ford 2011) clearly do have an advantage. In Britain, a French immigrant was perceived as more integrated than immigrants from other countries (with Muslims being worst off). Although the immigrants from 'preferred' backgrounds are not exempted from any particular integration criteria and their efforts on each one of them is rewarded as much as if they were Muslim, just being white or from a group with a good reputation in their particular host country makes them look more integrated from the start. While we think one explanation for this finding is that people use origin as a cognitive shortcut by which they reward members of immigrant groups that are portrayed as less 'problematic' on the group level, we think that our results also signal that a significant level of racial prejudice still plays a role in the differential perception of immigrants (although this finding needs testing for the Netherlands). We see that people holding ethnocentric attitudes are more likely to perceive a black immigrant as less integrated on average than they do a white immigrant. While integration is broadly a matter of consensus, prejudice towards certain origins is not.