Polarization in a consensual multi-party democracy – attitudes toward immigration in Norway

ABSTRACT

This paper studies polarization of attitudes toward immigration in Norway from 2001 to 2019. The paper studies polarization along five dimensions: dispersion, bimodality, consolidation, constraint, and sorting. Empirical analyses were based on two Norwegian longitudinal, cross-sectional surveys. The findings suggested that, first, overall attitudes toward immigration did not become more polarized in terms of dispersion and bimodality. There was, however, a tendency toward increased polarization of attitudes toward Islam and a decreased polarization of attitudes toward refugees. Second, there was an increasing generational gap in attitudes toward immigration, especially with respect to Islam. Third, attitudes toward immigration were more closely linked to attitudes toward other political issues and to party preference. Although these changes should not be overestimated, finding increased tendencies of consolidation, constraint, and sorting in a consensus-based democracy like Norway indicates the wider existence of polarizing trends similar to those in the UK and US.

Introduction

It is a commonly held view that public debates and party politics, especially in the US, have become more polarized (McCarty, Poole, and Rosenthal 2016; Sunstein 2017). Several US-based studies have shown that, in particular, affective polarization – i.e. the dislike of political opponents – has increased over the years (Iyengar et al. 2019; Druckman et al. 2020). The public’s deeply partisan responses to calls for COVID-19 containment efforts and vaccination are a testament to the fact that polarization in the US is not merely confined to political elites with mutual distrust and contempt, but that the US population is currently profoundly divided along party lines.

Increased affective polarization does not need to correspond with increased attitude polarization, and the literature is more mixed when it comes to the latter (Hetherington 2009; Adams, Green, and Milazzo 2012; Fiorina 2016; Gentzkow 2016). Furthermore, the dynamics of polarization in bimodal party systems cannot be readily transferred to multi-party settings with a long history of elite compromises. The evidence for recent changes in Scandinavian multi-party systems is somewhat scant (but see Jensen and Thomsen 2013; Aasen 2017; Reiljan and Ryan 2021). In this article, we contribute to the filling of this gap in empirical literature by exploring the development of attitude polarization in a Scandinavian context.

Compared to most other Western democracies, Norway is (still) relatively egalitarian and homogenous and has historically been characterized by a political consensus culture (Stenius 2010). While ideological “sorting” of voters is now increasingly taking place in the US, in the Norwegian multi-party system, which reflects century-old political cleavages resulting from critical junctures in history, voters have already been “sorted” by their geographical, cultural, and economic positions. Mutual disdain and conflicts of interest could be fierce among subcultures. Nonetheless, cross-cutting cleavages and elite compromise through coalitions have counteracted deep-seated conflict in the population (Rokkan and Lipset 1967). Norway is also characterized by comparatively high levels of political and social trust (Delhey, Newton, and Welzel 2011; Wollebæk et al. 2012), which should act as a bulwark against several forms of polarization.

Until quite recently, a lack of polarization has been seen as a greater problem than the opposite. Rewinding back a few years, electoral research in Norway showed that a majority of voters perceived the differences between different political parties to be small (Narud 2007). To be sure, this shift in perspective is not unique to the Norwegian case; in 1950, the main political science debate in the US was concern over too little polarization. In a special issue of the American Political Science Review, it was argued that the two main parties did not offer clear ideological alternatives, and in effect, voters did not have a real choice (APSA 1950).

However, stronger public concern regarding increasing polarization has emerged in recent years in Norway as well. New, contentious topics not accommodated by, and thus disrupting, the traditional party system have become salient. In particular, these topics include immigration and climate. The political developments in the US and UK have also impacted domestic discourse and fuelled debates on, for example, nationalism, globalization, and identity politics. The use of the word “polarization” in Norwegian media tripled from 2007 to 2017 (Prebensen 2018), and increasing polarization is frequently mentioned in concerned op-eds and public discourse.

However, perceptions of polarization do not necessarily reflect the de facto polarization of attitudes. Political actors, the media, and the general public may perceive the opinion climate to be more polarized than it actually is (Baldassarri and Bearman 2007). In this article, we contribute to the research field on polarization with a detailed study on the polarization of attitudes toward immigration in Norway. Have the attitudes of the population become more polarized, and if yes, on which dimensions of polarization are changes apparent? Are the trends the same across different aspects of immigration, including views on integration, refugees, and Islam? Do we see similar trends as those in the US and UK with more “sorted” identities and a growing generation gap in attitudes toward immigration?

Immigration was chosen as the point of study because it has been one of the most salient and contested political issues in recent years, both internationally and in Norway. It is at the nucleus of an emerging “transnational cleavage” deemed to be of similar importance to past critical junctures in European political history (Hooghe and Marks 2018). The issue has been identified as key in explaining the victory of Donald Trump in the 2016 US presidential election (Sides, Tesler, and Vavreck 2018). It was also the single most important issue among voters in the Norwegian parliamentary election in 2017 (Jenssen and Ivarsflaten 2019), while it was only ninth in 2001 (Bergh and Karlsen 2017). As explained in more detail below, this increase in the importance of this issue occurred simultaneously with the drastic increase in the share of the population being foreign born.

Based on two high-quality longitudinal data sources, we study trends in different measurements of polarization (DiMaggio, Evans, and Bryson 1996; Hoffmann and Miller 1997; Evans 2003). First, relying on a detailed survey on attitudes toward immigration and integration carried out approximately every other year since 2005, we examine changes in the dispersion and distribution of survey responses to items measuring different aspects of immigration and integration. Second, using the National Election Study in Norway, carried out every fourth year, we analyze the association between attitudes toward immigration and other ideological dimensions (constraint) and correspondence with party choice (sorting). Third, based on both data sources, we study how attitudes toward immigration are linked to background characteristics (consolidation).

The rest of the paper is structured as follows: First, we describe and discuss DiMaggio, Evans, and Bryson’s (1996) approach to polarization and spell out five dimensions for studying polarization as a process. Next, we describe the Norwegian context with regard to immigration and attitudes toward immigration, integration, and diversity. This is followed by a presentation of the data and methods, results, and discussion.

Dimensions of attitude polarization

Our approach to attitude polarization in this article builds on the work done by DiMaggio, Evans, and Bryson (1996). They viewed polarization as a multidimensional construct capturing the extremity and distance of different opinions. Following their analytical scheme, we examined the characteristics of attitude distribution (dispersion and bimodality), its relationship to background characteristics (consolidation), congruence with other political views, and party preference (constraint and sorting). We approached polarization as a process, i.e. changes in attitudes and not as a state, as well as the degree of polarization at a given time measured against a theoretical maximum.

DiMaggio, Evans, and Bryson (1996) distinguished four key elements of polarization. Below are these elements in relation to the topic of our study:

1. Dispersion: How dispersed are views on immigration? Opinions are said to be polarized when attitudes are spread out on a dimension and both sides are to some degree represented. The more dispersed the attitudes are, the more difficult it will be to maintain political consensus.

2. Bimodality: The public is considered polarized along this dimension if they cluster in separate camps with distance between themselves and other clusters. If opinions are concentrated in separate modes, social conflict is more likely to occur. If two positions are isolated from each other, it becomes harder for societies to find persons in in-between positions who can negotiate and act as the middle person. Bimodality can also increase spirals of silence (Noelle-Neumann 1974) and move politicians to mobilize based on more extreme positions rather than based on the median voter.

3. Consolidation: The public is polarized when attitudes toward an issue are highly correlated with relevant social characteristics, such as education, gender, age, income, or place of residence. The greater the association between attitudes toward an issue and salient individual characteristics, the more likely they will lead to social conflict. Furthermore, there will be polarization when differences between groups increase while differences within groups decrease, making them more homogenous.

4. Constraint: The public is polarized when their attitudes toward one thematic area, e.g. immigration, are highly correlated with attitudes in other areas. When attitudes toward immigration consistently follow attitudes toward, for example, climate or tax policies, the potential for conflict increases.

In addition, this work studies polarization along a fifth dimension, namely party sorting, which has gained increasing attention, especially in the US two-party system (Davis and Dunaway 2016; Fiorina 2016). Party sorting is related to constraint and describes the degree to which an individual’s ideological self-placement matches his or her partisan affiliation. On aggregate, increased party sorting leads to a tendency where social identities and attitudes increasingly follow party lines. Just as with consolidation described above, constraint and party sorting are likely to lead to fewer areas of agreement, making it harder to find compromise. Thus, polarization along these lines may be viewed in light of overlapping and cross-cutting cleavages (Rokkan and Lipset 1967). Rokkan and Lipset argued that cross-cutting cleavages – as opposed to overlapping cleavages – allow for more contact points and thus reduce conflicts. Overlapping cleavages resemble what has been referred to as culture wars, stacked identities, etc. in recent popular discourse (Klein 2020).

Thus, it is important to emphasize that this paper deals with attitude polarization and not the emotional intensity (affective polarization) attached to immigration attitudes (Iyengar et al. 2019; Druckman et al. 2020), or the degree of incivility in rhetoric and tone used in debates about immigration.

Although polarization is often portrayed as being disruptive to political processes by making it more difficult to find common ground and compromise, polarization, in the sense of distinct and diverging political positions, is not negative in and of itself. Rather, some degree of polarization is a democratic necessity; it clarifies alternatives to voters (Klein 2020) and helps the electorate hold the government accountable by raising electoral stakes (Testa 2012). Thus, a lack of polarization may also be a problem. Originally, the concept of political constraint came with positive connotations and was used as a measure of political sophistication (Converse 2006).

However, if polarization is increasingly coupled with (a) fragmentation of the population, expressed as increased constraint, consolidation, sorting, and mutual disdain between groups (Mason 2018), and (b) hostile rhetoric, it may have detrimental consequences for social cohesion and democracy. It may erode trust between people and in institutions, the ability to reach compromises, and the conditions of collective action (Cornelson and Miloucheva 2020). McCoy, Rahman, Somer (2018) argued that social cohesion is under threat when differences become aligned within camps with mutually exclusive, crystallized identities and interests, fired up by venomous political rhetoric. They postulated a causal link from polarization to democratic erosion in a cascading effect via aggressive “us and them” rhetoric centering on a single cleavage, deepening affective polarization, and ending with perceptions of the “other” as an existential threat.

Even though Scandinavian countries are a long way from the endpoint of this trajectory, it is still important to examine whether we see traces of such processes in changing attitudes. The present analyses are limited in the sense that they cannot address neither rhetoric and affective polarization nor the polarization occurring at the elite level. However, substantial changes across several of the dimensions outlined above would constitute a necessary (but insufficient) condition for the cascade of events described above to occur.

Past research on attitude polarization

Most extant research on polarization stems from the Anglo-American context. Conclusions regarding whether attitudes have become more polarized or not depend on conceptualization and measurement (Evans 2003; Fiorina and Abrams 2008; Hetherington 2009), as well as which issues are studied (Menchaca 2021). A recurring finding from the US is that the population’s political attitudes have become somewhat more polarized on some issues (race and morality) than others (economy), but the main change over time is that voters hold more consistent ideological beliefs that are increasingly sorted by their social identities and party preferences (Abramowitz and Saunders 2008; Fiorina 2016; Gentzkow 2016).

Studies have also indicated a growing generational gap. For instance, in recent elections in the US (Trump) and the UK (Brexit and 2019 General Election), generational differences in voting patterns seem to have increased (Sides, Tesler, and Vavreck 2018).¹

The evidence for recent changes in the Scandinavian context and its more consensus-based multi-party systems is scanter (but see Jensen and Thomsen 2013; Aasen 2017; Boxell, Gentzkow, and Shapiro 2020; Knudsen 2020). Polarization in Scandinavia appears less profound than in the US, but there is more similarity when analyzing its effects on voters holding opposing views (Knudsen 2020), especially among voters with strong partisan identities and more extreme political attitudes (Reiljan and Ryan 2021).

With regard to immigration, Swedish survey-based studies showed polarized attitudes toward refugees even before the refugee crisis (Ryan and Reiljan 2018) and asymmetric affective polarization between far-right voters and immigrant liberals, with the latter disliking the former more than vice versa (Reiljan and Ryan 2021). Studies have also suggested that labour market competition may polarize attitudes toward immigration (Kaihovaara and Im 2020).

In Denmark, researchers studied the effects of morally charged anti-immigrant rhetoric on polarization of attitudes. Relying on survey experiments, Simonsen and Bonikowski (2022), found that political moralization contributed to affective, but interestingly, not to attitudinal polarization. According to the authors, their findings from the Denmark may help explain the emergence of intense anti-immigrant politics while attitudes remain stable.

There has been limited research on the polarization of attitudes toward immigration in the Norwegian context. Studying longitudinal survey data dating back to the 1980s, Hellevik and Hellevik (2017) found that the population has become increasingly more positive toward immigration and multiculturalism. However, they did not study tendencies toward polarization and based their argument on only a few general survey items. The National Election Study has shown a remarkably stable picture over time when it comes to the attitudinal distribution of attitudes toward immigration, but immigration as a political issue has become more important to voters over time (Jenssen and Ivarsflaten 2019). Electoral research on ideological constraints in Norway suggests that attitudes toward immigration are correlated with attitudes toward the environment and economic left–right cleavage (Aardal, Bergh, and Haugsgjerd 2019).

The Norwegian context

Norway has a relatively short immigration history, starting with labour immigration in the 1970s, followed by refugees and asylum seekers from the mid-1970s onward. Family immigration became a factor beginning in the early 1990s (Brochmann and Kjelstadlie 2008). As a result of the extension of the European Union (EU) in 2002, European labour migrants from Eastern Europe (e.g. Poland and Lithuania) now constitute the largest immigrant groups in Norway. The share of foreign born increased from 6% in the year 2000 to 15% in 2020. In addition, in 2020, 4% of all children born in Norway were to two immigrant parents (SSB 2020).² The impact of this demographic shift on attitudes toward immigration has yet to be established (Craig, Rucker, and Richardson 2018).

Levels of interpersonal generalized trust as well as institutional trust are high and stable in Norway (Torcal 2016). However, recent research has shown that many did not consider immigrants as part of their “radius of trust” when responding to survey questions on generalized trust (Kumlin et al. 2017). Norwegian society is marked by a political consensus culture (Stenius 2010), a sentiment also mirrored in yearly negotiations between employers and labour unions, geared toward avoiding conflict and finding common ground.

Based on this contextual description, two competing hypotheses evolve. On the one hand, a consensual democracy with high levels of trust would lead one to hypothesize that attitudes toward immigration have not become more polarized over time. On the other hand, the relatively large demographic changes in the last few decades, combined with immigration being a more important political issue for voters (Jenssen and Ivarsflaten 2019), would lead one to hypothesize that attitudes toward immigration have become more polarized. These competing hypotheses will be studied empirically in the next sections.

Materials and methods

The analyses relied on two data sources: Five waves of The National Election Study, carried out from 2001 to 2017, and the Integration Barometer, carried out regularly in the period from 2006 to 2019. The two data sources have distinct advantages. The Integration Barometer contains a wide array of questions covering different aspects of attitudes toward integration and immigration, while the National Election Study captures views on other salient political issues as well as immigration. We used data from The Integration Barometer to study dispersion, bimodality, and consolidation, whereas data from the Election Study was used to study constraints of political positions and sorting, i.e. the correlation between attitudes toward immigration and other political attitudes and party preference. Both surveys revised the questionnaire for each new wave, and some items were removed and reinstated from round to round. Thus, our selection of items was limited to those included throughout the entire period of study.

Dependent variables (the Integration Barometer)

We created one composite index and three sub-indices with two items each covering attitudes toward refugees, attitudes toward integration, and attitudes toward Islam. All items were strongly correlated. A factor analysis (Table A2 in the appendix) showed that all items conformed to one dimension, explaining 55% of the variance with a Cronbach’s alpha of .83. Attitudes toward refugees were captured by the statements “Norway should receive more refugees in need of protection” and “My municipality should settle more refugees.” Attitudes toward integration were captured by “The relationship between immigrants and the rest of the population will improve” and “By and large, how well do you think the integration of immigrants in Norwegian society works?” Lastly, attitudes toward Islam were captured by “I am sceptical of people of Muslim faith” and “Do you think that the values within Islam are compatible with fundamental values in Norwegian society?” All statements except one were prompted by the question, “How well or poorly do the following statements fit with your opinion?” measured on a scale from 1 (“Does not fit at all”) to 4 (“Fits completely”). The exception is the question about how well integration works, which was measured on a scale from 1 (“Very poorly”) to 5 (“Very well”). The three sub-indices (refugees, integration, and Islam) were recoded so that 0 reflected a neutral position.

The items included in the indices were ordinal Likert-style variables. Treating these as continuous variables in the index constructions rested on the assumption that the distances between the ordinal values were the same, while they were in fact arbitrarily assigned. Therefore, in the construction of the main index, we employed categorical principal components analysis (CATPCA), which allowed variables to be scaled at different levels to make the modelling of nonlinear relationships between variables possible. A one-dimensional solution was requested from the CATPCA procedure in SPSS, and the saved score from the analysis was used as the basis for the analyses in the following. Imputation of missing values was used, but cases with missing values on more than two of the variables in the index (1.3% of cases) were removed from the analysis. The saved variable was very strongly correlated with a simple additive index of the seven items (r = .98). The subindices were constructed as simple additive indices.

Dependent variable (National Election Study)

We created one composite index (Cronbach’s alpha = .70) consisting of three items measuring attitudes toward immigration as a threat to national distinctiveness, immigrants’ right to social benefits, and general views on allowing immigrants to enter Norway. This index was also constructed using optimal scaling. The resulting index was strongly correlated (.97) with a simple additive index.

Six indices were constructed for the analysis of constraint. The left–right dimension was captured by the items “We should reduce state control over private industry,” “To exhort people to greater effort, we should be willing to accept bigger differences in wage levels,” and “It is more important to develop public services than to reduce taxation” (reverse coded) and a self-placement left–right scale (0–10). Anti-environment attitudes were measured by “To secure economic growth we need to develop our industry even if it conflicts with environmental considerations” and placement on an 11-point scale between “Environmental protection should not be taken so far as to affect our standard of living” and “More should be done for environmental protection, even if it means a considerable reduction in the standard of living for everybody, yourself included” (reverse coded). Christian values were measured by a single item: “We should promote a society where Christian values are more prominent.” Anti-globalization was reflected by disagreement with the item “We should go for a society that has a more international orientation, with less emphasis on boundaries between peoples and countries.” Centre-periphery was placed on a 0–10 scale between the statements “Central government pays too little regard to peripheral areas” and “Central government pays too much regard to peripheral areas” and the statements “Politicians and bureaucrats in Oslo don’t understand what’s really going on in the countryside in Norway” and “Big cities should keep significantly more of their own tax income than what they are presently allowed to” (reverse coded).

Since some of the items were assessed on 11-point scales and others on five-point Likert scales, all items were converted to z-scores before addition.³ Descriptive statistics of all items and indices used are included in the appendix.

We used various methods to analyze dispersion, bimodality, consolidation, constraint, and sorting (DiMaggio, Evans, and Bryson 1996; Hoffmann and Miller 1997; Evans 2003). Dispersion was measured by variance, indicating the extent to which data points differ from the mean. The quantification of bimodality is less straightforward, and each of the measures used in the literature has its limitations. Kurtosis, which was proposed by DiMaggio, Evans, and Bryson (1996), measures whether a distribution has heavy tails or outliers, with high values indicating absence of polarization. However, kurtosis fails to account for multimodal distributions (Downey and Huffman 2001). The bimodality coefficient (BC) is a standardized (0–1) measure that also takes into account the skewness of the distribution. Bimodality is present if the BC score is above .56 (Lelkes 2016). A limitation of BC is that it produces large coefficients also in highly skewed unimodal distributions (Pfister et al. 2013). Following the recommendations of Menchaca (2021), we opted for a nuanced approach examining both kurtosis and the BC, interpreting the latter as a matter of degree rather than a binary threshold indicating presence or absence of bimodality, combined with necessary visual inspections of the distributions.

Consolidation and constraint were measured using bivariate correlations and multivariate regression models. Bivariate correlations were used to measure the relationship between attitudes toward immigration, different sociodemographic characteristics, and attitudes toward other political issues over time, whereas multivariate regression models were used to estimate the total explained variance (r²) between all variables. Finally, sorting was measured using analysis of variance (ANOVA) models, which provided between- and within-group variation. The higher the between-group variation and the lower the within-group variation, the more polarized the attitudes were in terms of sorting.

Results

We first present an overview of the changes in attitudes toward immigration in the period under study before analyzing trends in polarization along five indicators: dispersion, bimodality, consolidation, constraint, and sorting.

Figure 1 shows changes in mean values on the composite index and the three subindices from the Integration Barometer. Overall, the composite index showed a slightly higher value in 2017–2019 compared to 2006–2009, suggesting a fairly stable but marginally positive trend in attitudes toward immigration in Norway over these periods. This contrasts with some other studies, suggesting that attitudes toward immigration over time in Norway have changed considerably in a more positive direction (Hellevik and Hellevik 2017).

Figure 1. Mean values of indices of attitudes toward immigration and integration.

However, considering the three sub-indices, we see some different trends in attitudes depending on the more specific topics. Attitudes toward Islam became more positive in the first period (2006–2012) and have since fluctuated. In contrast, changes in attitudes toward refugees were small between 2006 and 2013, but since then, attitudes have become more positive. We have no data points between 2013 and 2017, and one possible interpretation is that attitudes became more positive following the refugee crisis in 2015. The refugee sub-index is the only one in which there was a significant difference in attitudes in 2019 and 2006 (p < .001). Attitudes toward integration have been stable over time (except for a slight shift in 2012).

Dispersion and bimodality

Dispersion and bimodality were measured with variance and kurtosis. Figure 2 shows trends in these measures over time. Variance and kurtosis were interpreted as follows: The higher the variance, the longer the average distance is between randomly chosen respondents. The lower the kurtosis, the less normally distributed the data is. Thus, a negative kurtosis indicated heavy tails and more polarization in the distribution of answers.

Figure 2. Dispersion and bimodality of attitudes toward immigration and integration: variance and Kurtosis.

Overall, the results in Figure 2 indicate no overall tendency toward increasing polarization in terms of dispersion and bimodality. The composite index suggested a tendency toward bimodality (lower kurtosis) between 2006 and 2009, but increased kurtosis from 2009 onward. The difference between the variance in the first and last two waves was insignificant (Levene’s test, p = .373).

Using the bimodality coefficient (BC) as an alternative measurement of polarization (Lelkes 2016), the distribution on the composite index appears to be around the threshold of bimodality (.56). The BC was .55 in the first two waves and .57 in the two last waves. However, this is mainly due to the limitation of the BC noted above; it produces high values with highly skewed unimodal distributions (Pfister et al. 2013). On visual inspection, the distribution appears left skewed and unimodal in both the first two and last two waves (see Figure A1 in appendix).

Considering the sub-indices, there was a tendency of increased polarization in attitudes toward Islam (higher variance and BC, lower kurtosis) but decreased polarization in attitudes toward refugees. Changes in the variance from the first to last two waves of the survey were significant for both indices (Levene’s test, p < .000). There was only moderate changes in BC over time. According to this measure, attitudes toward Islam and refugees were both moderately polarized at a similar level (.46 for Islam and .47 for refugees and) in 2006–2008, while the coefficient was somewhat higher for Islam (.49) than refugees (.43) in 2017–2019. Attitudes toward integration had approximately the same level of polarization in the last two waves as in the first two, both in terms of variance (Levene’s test, p = .205), kurtosis and the BC (.38 in 2006–2008 and .39 in 2017–2018).

The opposite trends in the polarization of attitudes toward Islam and refugees are illustrated in Figure 3, which displays the distribution of each index in 2006 and 2019. The visual inspection of the distributions in Figure 3 indicated that Norwegians are moderately polarized on the issues of Islam and refugees, albeit not reaching the .56 bimodality threshold of the BC. Considering the attitudes toward Islam (left panel), the figure shows that the midpoint has weakened, indicating that fewer respondents held moderate positions in 2019. The opposite is true when considering attitudes toward refugees, where more people held moderate opinions in 2019 compared to 2006.

Figure 3. Distribution in attitudes toward Islam and refugees, 2006–2008 and 2017–2019 (%).

Consolidation

Next, we analyzed the relationship between attitudes toward immigration and sociodemographic characteristics. Figure 4 displays the total variance in attitudes toward immigration explained by gender, age categories (18–29, 30–49, 50–66, and 67 and over), education, household income, and country of origin.

Figure 4. Attitudes toward immigration and sociodemographic characteristics; explained variance (r²) from multivariate regression models with two-period moving averages.

Overall, there was no tendency toward increased consolidation. The background characteristics explained more or less the same proportion of the variance in the composite index, fluctuating between 7 and 10% and slightly declining after 2010. Attitudes toward Islam broke with this trend. The explained variance increased after 2013, suggesting that attitudes toward Islam have increasingly divided the population along sociodemographic characteristics. In 2006, background characteristics explained 7% of the variance, increasing to 17% in 2019. There was no such increase with regard to attitudes toward integration and refugees.

Women, specifically those residing in the capital Oslo and especially those with higher education, express more positive attitudes toward immigration compared to men, residents in other areas, and those with less education. Household income was generally uncorrelated with immigration attitudes. These patterns were stable over time.

The one exception to the general pattern of stability, however, is age. Since 2013, there has been an increased generational gap, in which older respondents expressed more negative attitudes toward immigration relative to younger respondents. Figure 5 shows the contrasting changes in the linear relationship regarding immigration attitudes based on age and the strongest correlate of such attitudes: education.

Figure 5. Attitudes toward immigration, age, and education: bivariate correlations with two-period moving averages.

While the correlation with education remained fairly stable and high (less so for attitudes toward integration than the other two topics), the relationship with age underwent a clear change between 2013 and 2017, a period covering the refugee crisis in 2015. For attitudes toward Islam, young age changed from a weak to fairly strong predictor of positive attitudes. The change was driven both by younger respondents becoming more positive and middle-aged and older respondents becoming more negative. With regard to attitudes toward refugees, young respondents shifted from being slightly more restrictive in 2013 to being slightly more lenient than older respondents in 2019.

Thus, a generational gap is emerging, particularly in relation to how Islam is viewed, and the timing of the change corresponds with the refugee crisis, which may have been differently received by older and younger generations.

Constraint

Constraint is the extent to which attitudes toward immigration are associated with attitudes toward other political issues. For this measure, we used the National Election Study 2001–2017. Figure 6 displays the correlation between attitudes toward immigration and other relevant ideological dimensions in the Norwegian electorate: economic left–right, centre–periphery, economic growth vs. environmental protection, Christian vs. secular values, and global vs. national values. The figure also displays total explained variance (r²) from regression models including all variables, with immigration attitudes as the dependent variable.

Figure 6. The relationship between attitudes toward immigration and attitudes toward other ideological dimensions, 2001–2017. National election study. Correlation coefficients and explained variance (r²) from multivariate regression models.

The results indicated there is a much closer link over time between attitudes toward immigration and attitudes toward other political issues. The explained variance increased from 21% in 2001 to 40% in 2017. The increase is particularly evident with regard to the economic left–right axis, suggesting that positive (negative) attitudes toward immigration are increasingly associated with left-leaning (right-leaning) attitudes toward economic issues. The one exception from the overall trend is the centre–periphery axis, which was found to be weakly related to attitudes toward immigration throughout the period.

Sorting

Finally, we measured the extent to which attitudes toward immigration were associated with party preferences. Table 1 summarizes results from separate ANOVA of the election data from 2001 to 2017, highlighting the correlation between party preference and attitudes toward immigration and between- and within-group variation.

Table 1. Party voted for and immigration attitude index using ANOVA.

	2001	2005	2009	2013	2017
Correlation (eta) between party voted for and immigration attitudes	.38	.41	.49	.49	.60
Between-group mean square	30.85	40.19	49.95	40.48	75.74
Within-group mean square	.93	.95	.89	.83	.77
F-value	33.36	42.52	55.91	48.63	98.90
n	1624	1679	1451	1398	1620

The results suggest a strong increase in sorting by party preference, with an increase in correlation coefficients from .38 in 2001 to .60 in 2017. The increased correlation was caused by both increasing differences between the party voters (between-group mean square) and less internal variation within each party (within-group mean square). The tendency was equally clear if the anti-immigration Progress Party was removed from the analysis; eta increased from .25 in 2001 to .45 in 2017.

In other words, on this particular dimension – attitudes toward immigration – the various parties’ voters have become more homogeneous, and the differences between the various’ parties’ voters have increased. This trend is not limited to the party for which immigration is the most important issue but occurs across the political landscape.

Discussion

Based on important societal characteristics and recent demographic and political changes, we offered two competing hypotheses regarding changes in the polarization of attitudes toward immigration in Norway. On the one hand, from a comparative perspective, Norway is characterized as having a homogenous population, low inequality, and high levels of trust, which are all factors pointing to a direction of limited polarization. On the other hand, Norway has witnessed large demographic changes in recent decades, and immigration has become one of the most salient political issues. These trends point toward more polarization in attitudes toward immigration.

Following the empirical strategy offered by DiMaggio, Evans, and Bryson (1996), we studied changes in the polarization of attitudes toward immigration along five lines: dispersion, bimodality, consolidation, constraint, and sorting. With regard to change over time, the results suggested that attitudes toward immigration have not become more polarized with regard to dispersion (increased variance) and bimodality (clustering) in Norway, but there have been tendencies of polarization with regard to consolidation (generational gap), constraint (correlation with other ideological dimensions), and sorting (party differences). Furthermore, when disaggregating attitudes toward immigration, the results suggested that attitudes toward Islam have become more polarized across all dimensions, whereas attitudes toward refugees have become less polarized. Considering the magnitude of polarization, attitudes toward Islam and refugees appeared moderately polarized, while this was so to a lesser extent with regard to attitudes toward integration.

Regarding attitudes toward Islam, the association between immigration and Islam has become more prominent in Norwegian public debates over the past two decades. The majority of asylum seekers arriving in and around 2015 were from Muslim countries (Syria and Afghanistan). The organized anti-immigrant organizations (e.g. Stop the Islamization of Norway [SIAN]), alternative media platforms, and blogs (e.g. Human Rights Service/rights.no, resett.no, and Document.no) have a wide social media reach and devote most of their energy to criticizing Islam. Hatred of Muslims was also the main motivation for the most significant terrorist attacks in Norway in recent years (attacks on the government building and a political youth camp on Utøya on July 22, 2011, and an attack on a mosque in Bærum on August 10, 2019). As the debate on Islam has taken centre stage both in the immigration and integration debates, it is not surprising that it is in this area that we found the strongest indicators of polarization. There has also been a tendency toward stronger dispersion and bimodality in views and a growing generational gap. The youngest cohort, those born after 1991, appeared more positive toward both refugees and Islam, and in the case of Islam, has increasingly distanced themselves from older cohorts. A cross-sectional survey carried out in Norway, Sweden, and Denmark in 2015 confirmed this generational gap in all three countries (Lundby et al. 2017). One interpretation of this pattern may be that young adults have lived their lives in a more diverse Norwegian society. This has exposed them to and brought them into contact with immigrants and their descendants, including persons from Muslim backgrounds to a larger extent than the older generations.

The reduced polarization in attitudes toward refugees may be associated with the war in Syria and the resulting mass influx to Europa. This asylum crisis may have increased the overall understanding of the needs of refugees among the Norwegian population. However, the reduced polarization may also be understood in the context of an increasingly restrictive asylum and refugee policy in Norway during the 20-year period (Brekke, Røed, and Schøne 2017). This could contribute to bringing those critical to immigration more in line with those with positive views on the matter.

Perhaps the most consistent general trend in the analysis is the increased association between attitudes toward immigration and other ideological dimensions and party preferences. Attitudes toward immigration have amalgamated with other political cleavages. When this occurs, it becomes more difficult to find common ground and reach compromises, and its conflict potential increases.

On the one hand, this reflects trends in other countries, especially the US, where several studies have shown increasing polarization between the two large parties (sorting), their main issues (constraint), and their voter groups (consolidation) (Fiorina 2016; Klein 2020). On the other hand, this finding may also reflect the fact that immigration has become a more salient political issue for voters over the past decades, as it was rated as the most important topic during the 2017 general election (Jenssen and Ivarsflaten 2019). This has mainly been a mobilizing issue for the far-right Progress Party, but today, all political parties have a solid immigration platform and seek to distinguish themselves as more or less progressive/restrictive in their immigration and integration policies.

Polarization among political parties is not necessarily a bad thing to the extent that political alternatives appear clearer to the electorate (cf. APSA 1950; Narud 2007), but increased distance in policy positions between parties may be a reflection of underlying conflicts in the electorate.

Given the somewhat mixed results depending on the particular measure of polarization and the sub-topic issue studied in this article, it seems clear that changes in the polarization of attitudes toward immigration in general over time should not be overstated in Norway. However, some of our findings suggest that the consensual, high-trust political culture of Norway is not immune to polarization and that international trends of generational gaps, amalgamation of political cleavages, and party sorting are also relevant in this context. In particular, the topic of Islam has the potential to become an even more polarizing issue as the moderate middle is weakened, generations increasingly disagree, and developments in technology and media give anti-Islam voices much greater reach and attention than was the case before. As anti-Muslim sentiments merge with other political views over time, the potential of an escalating, virulent conflict with Islam at centre stage is certainly a concern.

This study has focused on mass polarization on immigration attitudes in a consensual multi-party system. Further studies considering other polarizing issues, such as climate, as well as less polarizing issues, in order to present a broader picture of trends in polarization in consensual multi party systems. We also call for comparative studies to elucidate the relative magnitude of changes in polarization over time and across political systems.

Data availability statement

The data sets used in this article are deposited at the Norwegian Centre for Research Data and are available upon application (www.nsd.no).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by Institutt for samfunnsforskning [grant number NORPOL project].

Notes

1 In the UK 2019 general election, post-election polling suggested that age seems to be a more important dividing line than social class; see https://yougov.co.uk/topics/politics/articles-reports/2019/12/17/how-britain-voted-2019-general-election.

2 https://www.ssb.no/befolkning/artikler-og-publikasjoner/oppdaterte-tall-om-innvandrere--42215

3 The combination of Likert and 11-point scales via z-score transformation was used in one analysis (Figure 6). In order to check whether the combination of different scales could have affected the results, we compared the results in Figure 6 with separate regression analyses including (1) recalculated indices with the three 11-point-items removed and (2) the three 11-point items as independent variables. In analysis (1), R² increased from .19 in 2001 to .34 in 2019. In analysis (2), it increased from .12 in 2001 to .28 in 2019. Thus, the tendency toward increased constraint reported is robust to the differences in scales in the items used.

Appendix

Figure A1. Distribution on composite index 2006–2008 and 2017–2019.

Table A1. Descriptive statistics.

	Min	Max	Mean	Std. dev.	n
National Election Study (2001–2017)
We should reduce state control over private business	1	5	1.93	1.31	9154
We should accept larger wage differences than today	1	5	1.66	1.39	9439
It is more important to expand public services than to reduce taxes (r)	1	5	2.17	1.28	9407
Self-placement left-right scale	0	10	5.46	2.23	9121
Left-right dimension index (high value = right)	−7.5	6.1	0.02	2.79	8858
We still need industrial development to secure economic growth	1	5	2.59	1.23	9303
(0) Environmental protection should not be taken so far that it reduces our standard of living vs. (10) Pursue much more environmental protection, even if it means a significantly lower standard of living for everyone – including yourself ®	0	10	4.27	2.21	9454
Environment index (high value = anti-environment)	−3.8	3.9	0.09	1.63	9262
We should go for a society with more weight on Christian values	1	5	1.75	1.40	9275
Christian values (z-score)	−1.5	1.3	0.07	0.99	9275
We should go for a society with a more international orientation (reverse coded)	1	5	2.28	1.27	9161
Anti-globalization (z-score)	−1.4	1.8	0.00	1.27	9161
(0) Central authorities pay too little attention to rural Norway vs. (10) Central authorities pay too much attention to rural Norway (reverse coded)	0	10	6.13	2.28	9364
Politicians in Oslo understand little of what is happening in rural Norway	1	5	2.89	1.13	9229
The big cities should keep more of the tax revenue (reverse coded)	1	5	2.81	1.31	9128
Center-periphery index (high value = periphery over center)	−4.4	6.6	−0.01	2.14	8843
Integration Barometer
By and large, how well do you think the integration of immigrants in Norwegian society works?	1	5	2.62	0.94	13290
The relationship between immigrants and the rest of the population will improve	1	4	2.63	0.72	13311
Integration attitudes index (high values positive)	−3.5	3.5	−0.25	1.40	13121
Norway should receive more refugees in need of protection	1	4	2.60	0.92	13249
My municipality should settle more refugees	1	4	2.26	0.93	13140
Refugees attitudes index (high values positive)	−3	3	−0.26	1.64	12984
I am skeptical of people of Muslim faith (r)	1	4	1.48	0.95	11348
Do you think that the values within Islam are compatible with fundamental values in Norwegian society?	1	4	2.22	0.90	12233
Islam attitudes index (high values positive)	−3	3	−0.26	1.64	10317
Composite index	−3.12	2.03	−0.02	1.03	13517

Table A2. Factor analysis. Pooled data from the Integration Barometer.

Component	1
By and large, how well do you think the integration of immigrants in Norwegian society works?	.612
The relationship between immigrants and the rest of the population will improve	.704
Do you think that the values within Islam are compatible with fundamental values in Norwegian society?	.764
I am sceptical of people of Muslim faith	−.734
Norway should receive more refugees in need of protection	.789
My municipality should settle more refugees	.831
Eigenvalue	3.306
Percentage of total variance	55.106