The dataset we employ was composed for The Living History Forum (Forum För Levande Historia) by Statistics Sweden (Statistiska centralbyrån, SCB) through a survey taken among adolescents in 473 classes in schools throughout Sweden in 2014. The population for the study consisted of adolescents in Sweden in the last year of primary school (Årskurs 9) and years 1–3 of secondary school (Gymnasiets Årskurs 1–3). The normal age range for these school years is 14 to 19, but due to the possibility of starting school one year early, and finish secondary school later the sample also contains 16 respondents that are 13 or 20–21 years old. The final sample consisted of 6664 respondents.Footnote 1 For the purpose of the analysis, respondents who had themselves immigrated, and respondents with two immigrant parents are not included, thus further reducing the sample to 5220.
In addition to the information gained from the survey, the dataset also includes individual information from the School Registry (Skolregisteret), Pupil Registry (Elevregisteret), and Population Registry (Register över totalbefolkningen, RTB), all with the respondents’ permission (see Statistics Sweden, 2014). In addition to these sources, we include municipality-wide data from Statistics Sweden’s demographic database and from the Swedish Labor Office (Arbetsförmedlingen). A detailed table of descriptive statistics can be found in Appendix.
Use of the Term “Immigrant”
A pilot study was conducted prior to data collection where respondents’ interpretations of questions and terms used in the survey were analyzed. Terms like “immigrants,” in particular, may be interpreted to mean different things. As pointed out by Bevelander and Otterbeck (2016), “immigrant” is often used in a broader way than simply “a person who has immigrated to Sweden.” Everyday use of the term “immigrant” often includes people of immigrant descent more generally and reflects popular ideas concerning appearance, religion, culture or other identifying traits. Our pilot study found that “immigrant” is a salient social category for our responding adolescents and that this broad interpretation of the word corresponded with their understanding of it. The proportions of immigrants in our respondents’ municipalities and school classes are calculated based on the number of individuals who have themselves immigrated or who have two immigrant parents according to the national registry (RTB). Immigrants in Sweden are a highly diverse group and hail from all parts of the world. At the time of data collection, about 16.5% of the population in Sweden were born in a different country. Of these, 32.5% had an Asian background, 20% were from EU28 countries excluding the Nordic countries, 15.5% came from the other Nordic countries, 14.5% from non-EU European countries, 10% from Africa, 4% from South America, 2% from North America, and the remaining 1.5% from other regions or of unknown origin.
Selection and Operationalization of Variables
The dependent variable in this study is the level of negativity in attitudes toward immigrants, hereafter “negative attitudes towards immigrants.” This variable is measured as an index ranging from a low of 0 up to the highest level of negative attitude, indicated by 1. The index is composed of four statements, with a Cronbach’s alpha of .839. Respondents were asked to state how much they agree or disagree with these statements on a five-point scale:
Most immigrants are probably nice people.
It would be okay to live next door to an immigrant.
Immigrants cannot be trusted.
There are too many immigrants in Sweden.Footnote 2
These assertions tap into a number of dimensions of attitudes toward immigrants — including the acceptance of proximity, prejudices and preconceptions about immigrants — and both the cognitive and affective dimensions of attitudes, in line with other scales for measuring them (eg. Henry and Sears, 2002). Very importantly, these statements were also used in two previous questionnaires and the results are revealed in three reports on attitudes among Swedish adolescents (Lövander, 2010; Ring and Morgentau, 2003; Severin, 2014). The responses can be seen in Table 1.
The Explanatory Variables
Three different variables measure contact in this study. The first measure of contact is whether or not the respondent has friends who are immigrants and, if so, how many. Two more variables that measure contact were also created. The first of these uses the proportion of immigrants in the respondent’s school class as a proxy for contact and the second the proportion of immigrants in the respondents’ local area — measured as the municipality in which the respondent’s school is located. A common measure of contact with immigrants is their relative number within a given geographical area, often the municipality, town, or city (Wagner et al., 2006). While proximity is a proxy for contact — making contact more likely — a common critique has been that the measure does not necessarily detect actual contact nor capture its nature.
Because of this, we use three different variables for contact, so that we will have a more reliable measure than any of the three variables would offer alone. The three variables cover contact that occurs in different settings and contact that is qualitatively different. While the special importance of friendship as a type of contact has been established in previous research (Davies et al., 2011), there continues to be calls for additional research that accounts for the social context (Christ and Wagner, 2013; Pettigrew and Tropp, 2011), as well as contact on multiple levels of analysis. This may allow for more-detailed conclusions about the significance of contact at different levels of analysis.
In this paper, we also distinguish between low- and high-quality contacts. Low-quality contact is likely to be superficial, short-term, adversarial, or involuntary. In short, this type of contact does not fulfill the conditions predicted by Allport (1954) to be necessary in order for contact to reduce negative attitudes. High-quality contact, on the other hand, fulfills all or almost all of Allport’s optimal conditions and is amicable, voluntary, and happening under conditions of cooperation or common interest. Friendship is by definition high-quality contact but previous research has also pointed to contact which has friendship potential, as defined by Pettigrew (1997). Gaining friends is a process that takes time, and this research is cross-sectional in nature, so we think it important to acknowledge that amicable contact with friendship potential could occur that would not be captured by measuring friendship. A longitudinal research design might be able to capture this gradual development. However, as is also pointed out by Pettigrew (1997), the reduction of negative attitudes from contact is also a gradual process, so we believe that using the friendship variable to measure high-quality contact is well warranted. Individuals’ positive contact beyond friendship should be captured by the two other variables.
As outlined in the theory section, a measure of friendship with immigrants is a particularly critical variable for the robust measurement of the cross-group contact, particularly given the importance of social life among adolescents and the effects of these social processes for attitudes towards immigrants. In this study the concept of friendship is operationalized through respondents answering a question about how many friends they have who are immigrants, selecting from five alternatives: no immigrant friends; one immigrant friend; a couple of immigrant friends; many immigrant friends; or “I do not know.” This variable has been recoded into a series of four dummy variables with “no immigrant friends” as the reference category.
Contact in School
Contact in school offers a measure of long-term proximity to out-groups. However, contact in this situation is not necessarily of high quality, as the school setting can be competitive and it can potentially be where social conflict is played out. There may be only superficial contact or, according to past research, sometimes predominantly negative relationships between majority and minority pupils that, through generalization, lead to more-negative attitudes toward the entire minority group (Gieling et al., 2014; Stark et al., 2015). In previous research, schools were the setting where diversity had the most often been found to lead to more-negative attitudes (Thijs and Verkuyten, 2014). Regardless, measuring contact in school has the advantage that pupils have limited capacity to choose with whom they share a class and, hence, the problem of self-selection bias is reduced. In this study, the variable for contact in school is measured through the inclusion of the proportion of immigrants in the respondent’s class.
Contact in Local Area
The variable for contact in the local area is expressed as the proportion of immigrants in each municipality and is thought to increase the likelihood of all types of interaction with immigrants (Wagner et al., 2009). Over and above the other two contact variables, this adds to the analysis the everyday interactions and casual contact opportunities which may occur with other members of the local community.
The use of proportions of immigrants as a proxy for contact is a source of uncertainty that requires some attention. As has been pointed out in earlier work, for example by Pettigrew and Tropp (2011), the proximity to a minority that the proxy indicates does not necessarily mean that there is any actual contact. This may pose a particular problem when the geographical area studied is large. In such cases, there is room for significant internal differences and the possibility that processes of spatial and social segregation will prevent contact and meaningful interaction between and among groups. However, in this study, the purpose of this variable is to capture contact that is not friendship and that does not occur in school. This is likely to be superficial contact, so it is not our interpretation that this variable would capture anything beyond proximity to or superficial contact with immigrants.
To minimize the risk that proximity may not indicate contact, we include municipalities, as they are the smallest geographical units for which data exist. It is, however, hard to completely satisfy any potential uncertainty. The correlation measure between the number of immigrant friends and the proportions of immigrants in the local area and school class show, respectively, rs = .257 and rs = .287. These relationships suggest that non-immigrant adolescents in areas with more immigrants also form more friendships with immigrants — in other words, that opportunities for contact actually can be associated with more actual contact. We do not believe that the type of proxy variable is problematic when it is used as a complement to other contact variables and when one does not overstate its potential to represent actual contact.
The control variables included are factors identified in previous research on the contact–attitude nexus. The full list of control variables is as follows: gender, living in a big city, parents’ education level and occupational status, level of education/school track, immigrant background, and unemployment level in the municipality.
Gender and level of education are included because of their significant influence. Previous results for the effect of gender on attitudes are somewhat mixed, in general. Girls have usually been found to hold less-negative attitudes toward immigrants (Dixon, 2006; Hello et al., 2004; Ponce, 2017). A higher level of education is fairly consistently correlated with more-positive attitudes towards out-groups, although the strength of the correlation varies between the countries studied (Coenders and Scheepers, 2003). This has been found to be the case across Europe and North America (Hjerm, 2001) where evidence indicates that it is the education in itself that is the predominant cause, not competing explanations like socially desirable responding (Hainmueller and Hiscox, 2007; Vogt, 1997; Wagner and Zick, 1995). Whether the respondent lives in a large city or in a more-sparsely populated area is also controlled for with a dichotomous variable. The central municipalities of Sweden’s largest cities — Stockholm, Gothenburg, and Malmö — are differentiated from the rest of Sweden. In so doing, we account for differences in attitudes between residents of central urban areas and those in more-sparsely populated places and smaller towns and cities. This is a distinction that has been observed in previous research by Bevelander and Hjerm (2015) and Gorodzeisky and Semyonov (2009). Age often has a significant association with negative attitudes, though the variation in age in the surveyed population is so small that we do not include it in the statistical analysis. Moreover, age correlates problematically with the school track variable. It was however tested in a non-reported model and found to have no significant influence. Education level is measured through school track. A significant difference in attitudes towards various out-groups has been observed between pupils in vocational school tracks and pupils on university y preparatory tracks (Gniewosz and Noack, 2008; Lövander, 2010; Severin, 2014). Immigrant background is controlled for through the respondents’ parents’ immigrant origins. This variable distinguishes between respondents with two parents from Sweden and those who have one immigrant parent. Beyond being a control variable for the respondents’ background, having one immigrant parent is also an indicator of long established contact with immigrants. So this variable is also of interest as a form of contact, and we would expect a mixed background to show a negative association with the attitudes variable.
Respondents in this study are still dependents of their parents. Thus, their socio-economic background in the form of their parents’ educational level and occupational status is used as a measure of socio-economic status. This approach is also in line with research done in the past concerning prejudice towards other out-groups (Bevelander and Hjerm, 2015). To further emphasize the importance of accounting for socio-economic background, the transmission of attitudes from parent to child is an important explanatory factor. Gniewosz and Noack (2015) found such a transfer of attitudes to be strong up to about the age of 16, which is the median age for respondents in the present data.
Research has shown that levels of unemployment are connected to negative attitudes and aggression towards immigrants. Coenders and Scheepers (1998, 2008) found that population cohorts who experienced high unemployment in their formative years exhibited more support for ethnic discrimination. Thus, adolescents who live in areas with a high level of unemployment may be in competition with immigrants over entry-level jobs, which are thought to lead to an increased sense of economic threat, as has also been found in other previous research (Hjerm and Nagayoshi, 2011). In the statistical models, unemployment is the percentage of unemployed persons as measured by the Swedish Labor Office.
Methods and Limitations
We conducted two sets of regressions. We use models 1 and 2 to answer the first three questions of this study. The fourth and last question also relies on the results of models 3 and 4. The ordinary least regression method is used to measure the association between contact and attitudes toward immigrants, while controlling for individual and structural background characteristics. The second pair of regressions is performed using fixed-effect models to account for unobserved heterogeneity, serving as a robustness check and evaluating the two models tested previously. Furthermore, if the effect size of the variables decreases significantly when fixed effects are used, this would also indicate that the relative importance of measures on these contextual levels is high as opposed to individual-level variables. Thus the last two models also contribute to distinguishing the relative importance of the three levels of measurement. We believe it is an advantage that we are able to include all variables of interest in the OLS models, whereas, if we used only fixed-effects models, we would not be able to see the effect sizes for all the contextual variables. With this combination of models, we balance the limitations of each approach and can see both the effect sizes of the contextual variables and the model fit of the models while, at the same time, controlling for the impact that the multilevel structure of the data has on the results. Beyond the inherent limitations of OLS and fixed-effects methods there are, of course, other limitations to the current methodological approach, with some stemming from the data themselves. Firstly, we can infer no causal relationships due to the lack of longitudinal data and analysis. Secondly, we cannot distinguish between contacts with immigrants from different regions of the world, which would have added a further interesting dimension to the analysis. Another limitation with using cross-sectional data for contact studies is the aforementioned self-selection bias that cannot be controlled for.
As a further robustness check, we have also conducted the same analysis with a number of alternative non-reported models. The first included immigrants in the analysis so that the full sample can be analyzed; the second excludes individuals who have responded “I don’t know” when asked if they have immigrant friends. The last controls for respondents’ immigrant background with a 7-value measure of which region of the world they come from. This analysis confirmed that the model we rely on here is the one with the best fit and that these changes to the analysis did not significantly alter the results for our variables of interest.