ABSTRACT
The number of immigrants to Western Europe has been increasing, with immigration the subject of much controversy in contemporary Europe. In this article, we investigate the relationship between the size of the immigrant population, how natives perceive this size, and their anti-immigrant attitudes. We use data from the 2002/2003 European Social Survey covering 17 Western European countries, and we find that as a rule Western Europeans think that the immigrant population in their country is much larger than it actually is. The perceived size of an immigrant population has an impact on anti-immigrant prejudices, but the real size does not. Like many authors before us, we find that education reduces prejudice. However, we also find that around 10 percent of the total effect of education is a result of educated individuals' more accurate views about size of immigrant populations. In other words, the effect of education on prejudice is in part mediated by size perceptions.
Introduction
The post-war reconstruction of Europe resulted in a prolonged period of economic growth, and from the 1950s onwards countries in Western Europe began to attract growing numbers of immigrants. With the economic recession of the early 1970s, the labour-importing Western European countries adopted more restrictive immigration policies, but large-scale immigration continued nonetheless, up to the present day. As a result of these prolonged periods of large-scale immigration, the majority of countries in Northern and Western Europe had acquired sizeable (and growing) immigrant populations by the 1980s. Somewhat later, this also happened in Southern European countries such as Greece, Italy, and Spain (Castles and Miller 2003) As Massey et al. (1998: 109) put it: ‘In a few short years, the foreign-born percentage of Europe has come to rival that of older immigrant-receiving societies such as the USA, Australia or Canada’.
While political and ideological responses to this new reality have varied, two important common features can be observed. Firstly, immigration became a prominent issue in the political agenda of all of the Western European countries. Secondly, worrying signs of rising xenophobia and the increasing support for extreme-right political parties have been observed in a number of immigrant-receiving countries (Pettigrew 1998b). The increased size of the immigrant populations is one of the main reasons for the political prominence of immigration, and possibly also for a rise in xenophobic reactions to immigrants. With regard to xenophobic reactions such as prejudice and discrimination, both the actual numbers of immigrants in the country and the natives’ beliefs about the size of immigrant population in the country might be of importance. American research shows that individuals tend to perceive the minority populations to be much larger than they actually are (Alba et al. 2005). If this also holds true for Europe, then both the increased size of immigrant populations and Europeans’ misperceptions of this size are of relevance for analyses of attitudes toward immigrants and immigration.
The purpose of this article is to investigate associations between the size of immigrant population, natives’ perceptions of this size, and negative attitudes toward immigrants. We base our analyses on data from the 2002 to 2003 wave of the European Social Survey, and we focus on four main research questions. First, how accurate were the natives’ perceptions of the size of immigrant populations in the countries included in the study? Second, how did size perceptions influence anti-immigrant prejudice? Third, how can we understand differences in size perceptions among our respondents? And fourth, is the commonly observed negative effect of education on prejudice at least partly mediated by more correct size perceptions of highly educated individuals? We use simple correlational statistics to answer the first question, and a clustering corrected Structural Equation Model (SEM) to answer the second, third and fourth questions.
Theory and research
Prejudice against ethnic minorities is a well-researched topic in the social sciences and we know from previous research that such prejudice is influenced by many different individual-level and ecological-level variables. At the individual level, people who occupy a vulnerable position in the system of social stratification (who are uneducated, poor, unemployed, etc.) are often found to be more prejudiced (e.g., Raijman et al. 2003). These findings are usually explained as relating either to a fear of competition from minorities (e.g., Semyonov et al. 2006), or to psychological processes, such as the displacement of personal fears and anxieties onto others (see Quillian 1995 for a summary). With regards to education, the tendency for educated individuals to be less prejudiced has also been explained by socialization theory, and by a focus on the difference made by the acquisition of knowledge and information that comes about because of education. With regards to socialization, the educational system is considered to be an important socializing agent in transmitting liberal values, and consequently reducing prejudice (e.g., Hello et al. 2006). When education is seen as a process of knowledge acquisition, the focus is more on the thesis that ‘ignorance causes prejudice’ (e.g., Coenders and Scheepers 2003), and education is assumed to reduce prejudice by dispelling the simplistic beliefs that are at the root of stereotyping and prejudice. At the group-level, the focus is on ecological variables that can help to explain prejudice in individuals. The dominant theoretical bases here are group-threat or group-conflict theories, and it is expected that levels of prejudice rise in keeping with increases in the size of the minority population and with bad or deteriorating economic circumstances in the societies that are analysed (Quillian 1995).
With regard to prejudice, few would disagree that the size of an immigrant population matters. Just how and why the size matters, however, is a much more open question. Group conflict theories suggest that the increased size of minority population leads to increases in prejudice and discrimination (e.g., Blumer 1958; Olzak 1992). The reasons for the increased hostility of the dominant-group members may be related to their perceptions of a competitive threat, to an increase of economic competition between the dominant and the minority group, or to a real or perceived shift in the balance of power between the groups.
Contact theory (Pettigrew 1998a; Dixon 2006) suggests the opposite, namely that an increase in the size of a minority population can lead to more contact and more harmonious relations between the dominant and the minority group. The most influential theoretical argument within this line of research is that of Allport (1979). Allport maintained that equal status contact between majority and minority groups in the pursuit of common goals may reduce prejudice. All things being equal, the increased size of the minority population enhances the possibility of contact between members of the minority and the dominant group (Sigelman et al. 1996).Under the right conditions, such contact can lead to a reduction in intergroup hostility and prejudice. However, some authors have questioned the straightforward ‘more minorities – more contact’ thesis. Taylor (1998) discusses the possibility that the increase in the size of the minority group might result in less contact with the majority population due to increased segregation, while Sigelman et al. (1996) propose that minority–majority contact can be expected to rise in keeping with the size of the minority population – but only until a certain ‘tipping point’ is reached, after which interracial contact can be expected to decline.
The vast majority of empirical studies on the association between minority size and prejudice have been conducted in the USA, and most of these studies support the conclusion that an increase in the relative size of a minority population (usually blacks) leads to increased levels of prejudice and discrimination (e.g., Quillian 1996; Taylor 1998; Semyonov et al. 2000). The results of the European studies generally find either that larger minority populations are associated with stronger ethnic prejudice and hostility (Quillian 1995; Scheepers et al. 2002; Semyonov et al. 2006; Schneider 2008), or produce non-significant findings (Evans and Need 2002; McLaren 2003; Semyonov et al. 2004). An exception is a study conducted in Germany by Ulrich Wagner and his associates that finds that larger minority size (in districts in Germany) is associated with less prejudice (Wagner et al. 2006). The authors also find that the negative effect of minority size on prejudice is entirely indirect, mediated by better opportunities for contact between majority and minority members that are a result of increased proportion of minorities.
When it comes to the perceived size of minority populations, less empirical research has been carried out, probably because of the lack of appropriate data (Semyonov et al. 2004). Semyonov et al. (2004) found that their German respondents grossly overestimated the size of the foreign population in Germany. While foreigners constituted 8.32 percent of the population in the German districts included in the study, the average estimate of the size of the foreign population as respondents perceived it was almost twice as high, equalling 15.75 percent. Perhaps the most interesting finding of this study is that the perceived size of the foreign population was associated with anti-foreign sentiments, while the real size was not. The authors comment on their finding as follows: ‘… the data suggest that it is not the actual size per se that prompts anti-foreigners sentiments, but it is socio-psychological construct – perceived size of the out-group population that is associated with anti-foreigners sentiments’ (2004: 696).
Contrary to the findings of Semyonov and his collaborates, Lahav (2004) found her Western European respondents to be fairly well informed about the actual size of immigrant populations in the Western European countries included in her study. Comparing respondent perceptions of where (i.e., in which country) most immigrants resided with the actual numbers of immigrants living there, she found a very strong correlation between the two measures. Using the same data that we do, Sides and Citrin (2007) found that respondents overestimated the percentage of immigrants in all of the 20 Eastern and Western European countries included in their study. With regards to associations between misperceptions of the size and anti-immigration attitudes, the authors decide not to include these misperceptions as a single predictor variable in their multivariate models. Instead, they decide to include a predictor based on a somewhat vaguely formulated question: ‘Compared to other European countries of about the same size as [country], do you think that more of fewer people come to live here from other countries’ (Sides and Citrin 2007: 486). Both these measures of the immigrant size and the interaction between them are included simultaneously in the models, and the results show that misperceptions about the percentage of immigrants do not have significant positive effects on levels of prejudice in respondents who believed that their country received fewer immigrants than other countries, but that they do have significant effects on the rest.
In the USA, perceptions of the sizes of minority populations are somewhat more systematically researched. There is a considerable amount of evidence that whites strongly overestimate the relative size of a non-white population (Nadeau et al. 1993; Nadeau and Niemi 1995; Gallagher 2003; Alba et al. 2005). Misperceptions about size are indeed quite substantial, often amounting to ‘fairly extreme distortions of national demography’ (Alba et al. 2005: 911). With regards to the association between misperceptions about the size of the minority and anti-minority attitudes, Alba et al. (2005) find that the larger misperceptions are associated with more negative attitudes toward racial or ethnic minorities.
When it comes to explaining why there are such misperceptions about the size of minority populations, the authors of American studies provide a fairly large range of explanations, and in what follows we will provide only a brief subset of these. Gallagher (2003) uses qualitative research techniques to shed more light on misperceptions about the minority size and his findings indicate that the media, residential segregation, racial stereotypes and perceived threats all contribute to misperceptions. Nadeau et al. (1993) refer to high levels of innumeracy in the general population as one of the main reasons for such misperceptions.
The multivariate analysis carried out by Alba et al. (2005), where misperceptions of the minority size are the dependent variable, reveals education, gender and age to have an impact on misperceptions. Not surprisingly, education proved to have a particularly strong effect on misperceptions, with educated individuals having more accurate views. Since the authors also find that individuals with more accurate perceptions of minority size tend to be less prejudiced, this opens an interesting question whether education reduces prejudice indirectly, by improving knowledge about minority size. Education has been known to reduce prejudice or a long time, and in the USA there have been educational programs specifically designed to reduce prejudice. The research on effectiveness of these programs has produced mixed results (see Nelson 2002, for a summary), and persistence of prejudice in spite of efforts to reduce it has raised questions about effectiveness of educational programs designed to reduce prejudice and hostility toward outgroups. However, as Alba and his colleagues point out, misperceptions about minority size lead to increased hostility toward minorities, and this hostility could be reduced simply by providing better information. It is also reasonable to assume that misinformation regarding other relevant facts about immigrant population can contribute to increase in prejudice. Quite obviously, educational system is particularly well-suited for dissemination of correct information about minorities.1 Thus, in the words of authors, their findings ‘highlight the frequently overlooked value of an old bromide against prejudice: education’ (Alba et al. 2005: 901). As we shall see later, there are several theoretical explanations for effects of education on prejudice, and Alba and his associates are focusing on one particular type of effect, namely the effect of education on prejudice that is mediated by perceptions of minority group size.
In spite of plausible argumentation, there are no tests for statistical significance of the mediated effect of education in Alba et al. (2005). In other words, we do not know if there are any statistically significant mediated effects of education. We also do not know how strong these effects are, both in absolute values and relative to the total effect of education. To answer these questions, one needs to deploy different modelling technique, with most obvious candidates being path-analytical models and Structural Equations Modelling (SEM). In this article, we shall use SEM to analyse statistical significance and strength of effect of education on prejudice that is mediated by minority size perceptions in our European sample.
Data and methods
The main source of the data for the study is the 2002–2003 wave of the European Social Survey (ESS).2 In addition, we make use of data from a number of different sources for coding country level variables. Following Sides and Citrin (2007), we use OECD (2005) data for measuring the real size of the foreign-born population in the countries covered by our analyses.3 For all of the countries included in the analyses except Switzerland, the source of data about unemployment rates is Eurostat (2006). The Swiss data were taken from United Nations (2005). The source of GDP data for all of the countries is CIA (2003).
All of the 17 Western European countries covered by the ESS data file are included in our analyses. The data is weighted in such a way that each country has an equal number of respondents. The equal size weight is combined with the design weight from the ESS file that corrects for the inequalities in selection probability that are caused by the sampling design. Our focus is on the majority group members, and excluded from the analyses are foreign-born, non-citizens, and individuals who belong to a minority ethnic group.4 That reduces our sample size from 33,186 to 29,013 individuals (real sample sizes).
Since an important part of our analysis includes complex causal relations as well as predictors of both size perceptions and prejudice, Structural Equations Modeling is the most appropriate analytic technique. As we shall see, most of our variables are measured at the individual-level, but some are measured at the country-level. Thus, a multilevel SEM was required. However, our pre-testing of the multilevel SEMs shows that all but the simplest ones do not converge. The most probable reason for the convergence problems is the low number of level-2 units (i.e., countries). We therefore employ a clustering corrected single-level SEM.5 The estimation is performed by Mplus (version 4) software that also allows for a correct application of the weights from ESS and the appropriate treatment of missing values (Muthen and Muthen 2006).
Individual-level variables
The Perceived size variable is constructed on the basis of this question from the survey: ‘Out of every 100 people living in [country], how many do you think were born outside [country]?’ Thus, the variable measures the respondent's perception of what percentage of the country's population is made up of foreign-born individuals.
The dependent variable Anti-immigrant prejudice is a latent variable composed of six indicators. The indicators are based on questions from the survey which deal with the influence immigrants have on different aspects of the country's economic and social life. All of the indicators are measured on a 0–10 scale, and have been coded so that 0 represents the most positive opinion about immigrants, while 10 represents the most negative. A brief description of the indicators is as follows: Jobs – measures the opinion about the influence immigrants on job creation in the country (0, create new jobs; 10, take away jobs); Welfare – the influence on the health and welfare services (0, positive influence; 10, negative influence); Economy – the influence on the country's economy in general (0, positive influence; 10, negative influence); Culture – the influence on the cultural life (0, cultural life enriched; 10, cultural life undermined); General – immigrants make the country a better or worse place to live (0, better place to live; 10, worse place to live); Crime – the influence immigrants have on the country's crime problems (0, crime problems made better; 10, crime problems made worse).
Education is usually one of the most important predictors of anti-immigrant prejudice (Vogt 1997). In order to get as good a measure of the respondents’ level of education as possible, we operationalize education as a latent variable composed of two indicators: Edu. yrs. measures the respondent's number of years in full-time education, while Edu. level measures the highest level of education the respondent has achieved, with values varying from 0, ‘incomplete primary education’ to 6, ‘second stage of tertiary education’. Since educational systems vary in the countries covered by the survey, the last indicator is based on a recoding of the national data into a common coding frame (Norwegian Social Science Data Services 2006).
The control variable Female is a dummy variable coded ‘1’ for females. The variable Urbanity measures the size of the place the respondent lives in (1, ‘a farm or home in the countryside’; 5, ‘a big city’), while Age measures the age of the respondent (in years). Two dummy variables tapping personal difficulties in the life of the respondent: being Unemployed (coded ‘1’ if the respondent answered that (s)he is either ‘unemployed and actively looking for a job’, or ‘unemployed, wanting a job but not actively looking for a job’); and Financial problems (coded ‘1’ if the respondent answered that living on her or his present income was either ‘difficult’ or ‘very difficult’). The respondent's position in the labour market has been operationalized as a set of dummy variables indicating the occupational category. Include in our models are the variables White collar, Skilled blue collar, and Unskilled blue collar. To avoid losing observations for individuals who had no occupation, we also include the dummy variable No occupation. The reference category is ‘Managers and professionals’.
Our dataset enables us to include within the models measures of the number of immigrants at a place of work. The number of immigrants at the work place is operationalized as a set of two dummy variables, both based on the question: ‘Do you have any colleagues at work who have come to live in [country] from another country?’. The dummy variable Several immigrant colleagues is coded ‘1’ if the answer category chosen by the respondent was: ‘yes, several’, while the dummy variable Few immigrant colleagues was coded ‘1’ if the answer was ‘yes, a few’.
Country-level variables
The Real size variable is a country-level variable constructed on the basis of the data from OECD (2005), and measuring the actual percentage of foreign-born individuals in the country. We calculate the numbers of immigrants in the same manner as Sides and Citrin (2007), by subtracting the native-born from the total number of residents (see footnote 36 in Sides and Citrin 2007 for more details).6
The measure of GDP per capita is a Purchasing Power Parity (PPP) estimate for 2002 measured in thousands of US dollars. We also compose two measures for levels of unemployment: (1) Mean unemployment 1998–2002 measures the average level of unemployment in the countries in the survey in the five years leading up to the data collection. We use the mean value of a five-year period in order to reduce the possible influences of random short-term fluctuations in levels of unemployment; (2) Change in unemployment 1998–2002: This variable is created to test Quillian's (1995) hypothesis that deteriorating economic conditions lead to increased levels of prejudice. The variable measures differences in unemployment levels in 2002 and 1998, and is calculated as the level of unemployment in 2002 minus the level of unemployment in 1998.
Results
Let us begin the analysis by calculating the correlation between the real percentage of foreign-born individuals in a country and the average values of the respondents’ perceptions of that percentage (in the same country). The correlation between the real and the perceived percentage is positive and strong (r=0.79), and the coefficient is highly significant (P<0.001). At first glance, this result would seem to support the findings of Lahav (2004) that show respondents to be knowledgeable about the size of the minority in their country. However, if we look at the country-by-country numbers that are presented in Table 1, we see that the average value of the perceived size is larger than the real value for all of the 17 countries included in the analysis.
. | Perceived size . | Real size . | Misperceptions1 (Percentage points) . | Misperceptions2 (Percentages) . | N . |
---|---|---|---|---|---|
Austria | 19.82 | 12.48 | 7.34 | 58.83 | 1623 |
Belgium | 22.42 | 10.68 | 11.74 | 109.97 | 1711 |
Denmark | 9.95 | 6.73 | 3.22 | 47.87 | 1828 |
Finland | 6.61 | 2.54 | 4.07 | 160.23 | 1876 |
France | 27.01 | 10.03 | 16.98 | 169.34 | 1665 |
Germany | 19.53 | 11.12 | 8.41 | 75.67 | 1742 |
Greece | 19.93 | 10.27 | 9.66 | 94.03 | 1605 |
Ireland | 14.57 | 10.38 | 4.19 | 40.32 | 1742 |
Italy | 17.65 | 3.93 | 13.72 | 349.13 | 1875 |
Luxembourg | 39.21 | 32.45 | 6.76 | 20.83 | 1118 |
The Netherlands | 23.21 | 10.10 | 13.11 | 129.76 | 1781 |
Norway | 12.07 | 7.33 | 4.74 | 64.70 | 1804 |
Portugal | 20.63 | 6.29 | 14.34 | 228.03 | 1758 |
Spain | 16.06 | 5.32 | 10.74 | 201.87 | 1790 |
Sweden | 19.91 | 12.01 | 7.90 | 65.76 | 1720 |
Switzerland | 27.12 | 21.55 | 5.57 | 25.85 | 1545 |
United Kingdom | 23.54 | 8.28 | 15.26 | 184.26 | 1709 |
. | Perceived size . | Real size . | Misperceptions1 (Percentage points) . | Misperceptions2 (Percentages) . | N . |
---|---|---|---|---|---|
Austria | 19.82 | 12.48 | 7.34 | 58.83 | 1623 |
Belgium | 22.42 | 10.68 | 11.74 | 109.97 | 1711 |
Denmark | 9.95 | 6.73 | 3.22 | 47.87 | 1828 |
Finland | 6.61 | 2.54 | 4.07 | 160.23 | 1876 |
France | 27.01 | 10.03 | 16.98 | 169.34 | 1665 |
Germany | 19.53 | 11.12 | 8.41 | 75.67 | 1742 |
Greece | 19.93 | 10.27 | 9.66 | 94.03 | 1605 |
Ireland | 14.57 | 10.38 | 4.19 | 40.32 | 1742 |
Italy | 17.65 | 3.93 | 13.72 | 349.13 | 1875 |
Luxembourg | 39.21 | 32.45 | 6.76 | 20.83 | 1118 |
The Netherlands | 23.21 | 10.10 | 13.11 | 129.76 | 1781 |
Norway | 12.07 | 7.33 | 4.74 | 64.70 | 1804 |
Portugal | 20.63 | 6.29 | 14.34 | 228.03 | 1758 |
Spain | 16.06 | 5.32 | 10.74 | 201.87 | 1790 |
Sweden | 19.91 | 12.01 | 7.90 | 65.76 | 1720 |
Switzerland | 27.12 | 21.55 | 5.57 | 25.85 | 1545 |
United Kingdom | 23.54 | 8.28 | 15.26 | 184.26 | 1709 |
Note: The Misperceptions1 measure equals the difference between the Perceived size and the Real size. The Misperceptions2 measure expresses misperceptions as a proportion of the Real size, and is calculated as: ((Perceived size – Real size)/Real size)*100.
N, number of respondents in the country (real sample sizes).
As one can see, there are sizable differences between the real and the perceived percentage of the foreign-born.7 If we judge misperceptions according to percentage points, the largest overestimations are found in France (17 percent) and the United Kingdom (15 percent). Although the populations of both these countries do in fact have quite substantial percentages of foreign-born individuals – comparable to levels found in the USA8 – their inhabitants seem to believe that around a quarter of the population in their countries is foreign-born. A possible explanation for the French and the British overestimates is the sizable presence of so-called ‘second generation immigrant’ populations in these countries, and it is possible that respondents also count these in their estimates of the foreign-born. However, if we look at the numbers for Italy and Portugal, we find gross overestimates in these countries too, even though they have much smaller foreign-born populations, and hardly any second generation ones.
If we express these misperceptions as a proportion of the real size (Misperceptions2 measure), we can see that in 8 out of our 17 countries, the respondents perceive the immigrant population to be more than two times larger than its real size. The largest misperceptions are observed in Italy, Portugal, and Spain – there the foreign-born population is thought to be more than three times larger than it actually is. We might also notice that, for some of our countries, the misperceptions are modest when expressed in percentage points. Danes thought the immigrant population to be only three percentage points larger than it actually is, and the perceptions of the Finns, Irish, and Norwegians are also less than five percentage points larger than the real values. Misperceptions that are a few percentage points off the mark might be considered reasonably accurate, but two things are worth noting. First, these misperceptions are large when compared with the real size of immigrant populations. Thus, the Danish estimate is still almost 50 percent larger than the real value, while the Finnish estimate is 160 percent larger. Second, although some are smaller than others, all of the misperceptions are in a positive direction. The random distribution of positive and negative misperceptions one might expect in the case of unbiased estimates is not present: the respondents seem systematically to overestimate the size of immigrant populations. To formally test if the overestimates are statistically significant, we conducted a one-sample t-test for each country in our data file. The t-test confirms that the respondents overestimate the size in all 17 countries (P<0.001 for all of the countries; two-tailed test; real sample sizes).
To sum up, then, two main conclusions can be drawn from the results so far. First, the respondents overestimate the size of the foreign-born populations in all of the countries in our sample – and for some countries, these overestimates are very considerable. Second, in spite of these large misperceptions, there is a ‘method in the madness’: for most countries there is a certain correspondence between the perceived and the real numbers. Thus, although both Britons and Finns grossly misperceive the size of their immigrant populations, the Finnish perceptions are still only a third of the British one, which is roughly proportional to the real sizes in these two countries. This explains the high correlation observed between the perceived and the real size, and probably also the divergent results of Lahav (2004).
The structural equation model
Let us now proceed with the multivariate analysis. Our three main variables of interest are the real and the perceived size of immigrant population, and the anti-immigrant prejudice. Hypothesized causal relations between these variables are presented schematically in Figure 1.
Hypothesized causal relations between the real and the perceived size of immigrant populations and anti-immigrant prejudice.
Hypothesized causal relations between the real and the perceived size of immigrant populations and anti-immigrant prejudice.
In addition to the three main variables, we also wish to include in our analysis variables known to influence prejudice and/or size perceptions. These are included in the full structural equations model, presented in Figure 2.9
Structural equations model predicting anti-immigrant prejudice and the perceived size of the immigrant population.
Structural equations model predicting anti-immigrant prejudice and the perceived size of the immigrant population.
Notice that one could also argue that it is possible that there is an additional causal effect from prejudice to perceived size, and that these two variables form a causal loop. This would lead to a more complicated ‘non-recursive’ structural equation model. We have tried estimation of such model, but there seem to be not enough country-level variation in our data, the standard errors of coefficients increase drastically in the non-recursive model. Since we did not find evidence of a strong positive influence of prejudice on misperceptions (the coefficient in the non-recursive model was actually negative), and since the dominating theoretical explanations point into direction of causal relations as in the Figure 1, we proceed with the analysis of a recursive model.
The model includes all of our level-1 variables, and two out of four level-2 variables were created taking group-conflict theories as the point of departure. Due to the low number of observations at level-2, the simultaneous inclusion of larger number of level-2 predictors would be problematic.10 Therefore, we conduct preliminary analyses with subsets of level-2 variables. Only Change in unemployment shows a significant effect on prejudice, and we therefore retain this variable in the final model. In order to further reduce the number of level-2 parameters estimated, we also remove the path from Change in unemployment to Perceived size, on the grounds that it did not have any significant effect in the preliminary analysis. The large number of causal links makes it difficult to present the values of the coefficients in the diagram. Thus, we only present loadings of indicators on latent variables, while the values of path coefficients and the indicators of fit are presented in Tables 2 and Tables 3.
. | Endogenous variables . | |||
---|---|---|---|---|
. | Anti-immigrant prejudice . | Perceived percentage . | ||
. | β . | t-ratio . | β . | t-ratio . |
Individual-level variables | ||||
Education | −0.226*** | −8.298 | −0.19*** | −10.261 |
Perceived size | 0.148*** | 6.941 | − | − |
Age | 0.007 | 0.392 | −0.067*** | −5.592 |
Female | −0.002 | −0.173 | 0.13*** | 10.218 |
Urbanity | −0.012 | −0.369 | 0.062*** | 6.079 |
Unemployed | 0.014 | 1.147 | 0.015 | 1.773 |
Financial problems | 0.097** | 3.199 | 0.065** | 3.24 |
Occupational category | ||||
White collar | 0.011*** | 0.587 | 0.011 | 0.927 |
Skilled blue collar | 0.07 | 4.069 | 0.015 | 1.102 |
Unskilled blue collar | 0.024 | 1.387 | 0.03 | 1.839 |
No occupation | 0.02 | 0.92 | 0.072 | 1.846 |
Immigrants at work | ||||
Several immigrant colleagues | −0.099*** | −6.043 | 0.031* | 2.436 |
Few immigrant colleagues | −0.06** | −3.285 | 0.012 | 0.971 |
Country-level variables | ||||
Real size | −0.104 | −1.792 | 0.331*** | 8.59 |
Change in unemployment 1998–2002 | 0.134* | 2.443 | − | – |
. | Endogenous variables . | |||
---|---|---|---|---|
. | Anti-immigrant prejudice . | Perceived percentage . | ||
. | β . | t-ratio . | β . | t-ratio . |
Individual-level variables | ||||
Education | −0.226*** | −8.298 | −0.19*** | −10.261 |
Perceived size | 0.148*** | 6.941 | − | − |
Age | 0.007 | 0.392 | −0.067*** | −5.592 |
Female | −0.002 | −0.173 | 0.13*** | 10.218 |
Urbanity | −0.012 | −0.369 | 0.062*** | 6.079 |
Unemployed | 0.014 | 1.147 | 0.015 | 1.773 |
Financial problems | 0.097** | 3.199 | 0.065** | 3.24 |
Occupational category | ||||
White collar | 0.011*** | 0.587 | 0.011 | 0.927 |
Skilled blue collar | 0.07 | 4.069 | 0.015 | 1.102 |
Unskilled blue collar | 0.024 | 1.387 | 0.03 | 1.839 |
No occupation | 0.02 | 0.92 | 0.072 | 1.846 |
Immigrants at work | ||||
Several immigrant colleagues | −0.099*** | −6.043 | 0.031* | 2.436 |
Few immigrant colleagues | −0.06** | −3.285 | 0.012 | 0.971 |
Country-level variables | ||||
Real size | −0.104 | −1.792 | 0.331*** | 8.59 |
Change in unemployment 1998–2002 | 0.134* | 2.443 | − | – |
Note: *P<0.05; **P<0.01; ***P<0.001 (two-tailed).
Fit measures:
Chi-Square=394.069; df=104; P-value<0.0001
CFI=0.952
RMSA=0.010
SRMR=0.021
From the standardized coefficients presented in Table 2 we can see that of all the predictors included in the model, education clearly has the strongest impact on prejudice. This confirms findings in countless other studies that also link education to strongly reduced levels of prejudice against minorities. Education also has a pronounced effect on the perceived size of foreign population, surpassed only by the effect of Real size. Since our respondents systematically overestimate the size of the immigrant population, the negative effect of education means that educated individuals have more accurate views.11 This is in line with findings in previous studies (Nadeau et al. 1993; Alba et al. 2005), and shows that education helps to reduce widespread misperceptions of size.
Perceived size has a positive effect on prejudice (it has the second strongest effect in our model, surpassed only by education). Interestingly, the real size does not have a significant effect on prejudice, and the coefficient is actually negative.12 This is precisely what Semyonov et al. (2004) found using German data, although in their study the different levels of measurement of the perceived and the real size did not allow for a proper specification of causal links (both variables were simply included into a SEM model as exogenous variables). The positive effect of perceptions of size on prejudice is in line with previous findings (Semyonov et al. 2004; Alba et al. 2005), and is easy to explain using group-threat theories. The absence of the positive direct effect of the real size on prejudice is, however, a less common finding. Real size has a strong positive effect on perceived size, and thus, a positive indirect effect on prejudice. Nevertheless, as we shall see, the total effect remains negative and non-significant. We also tried to remove the perceived size from the model, but the effect of the real size remained non-significant (and negative). Thus, the conclusion is that we do not find the positive effects of real size on prejudice that were observed in some previous European studies (e.g., Quillian 1995; Semyonov 2006).
When it comes to what influence larger numbers of immigrants at work have on anti-immigrant prejudice, having immigrant colleagues reduces prejudice. The effect of having immigrant colleagues can be interpreted in the light of the classical formulation of contact theory: prejudice is reduced by equal status contact in the pursuit of common goals (Allport 1979).
The variables related to a respondent's position in the system of occupational stratification have an effect on prejudice that is roughly in line with previous findings and theoretical expectations. All of the coefficients are in the predicted direction, although only Skilled blue collar has a statistically significant effect. The same is true for the variables which measure personal difficulties (Unemployed and Financial difficulties), although the non-significant effect of Unemployed is somewhat surprising. With regards to unemployment at the country level, the Change in unemployment has a positive effect on prejudice. This means that individuals from countries that have experienced a stronger rise in unemployment levels in the five years prior to the survey tend to be more negatively inclined toward immigrants. This is in line with group conflict theories, but the effects of the country-level variables should be interpreted with caution because of the small number of level-2 observations in our study.
Our exogenous variables can have indirect effects on prejudice as well as the direct effects via Perceived size. Let us now have a closer look on these effects. The total, direct and indirect effects are presented in Table 3.
. | Total . | Direct . | Indirect . | |||
---|---|---|---|---|---|---|
. | β . | t-ratio . | β . | t-ratio . | β . | t-ratio . |
Individual-level variables | ||||||
Education | −0.254*** | −9.075 | −0.226*** | −8.298 | −0.028*** | −5.021 |
Age | −0.003 | −0.186 | 0.007 | 0.392 | −0.01*** | −6.192 |
Female | 0.017 | 1.438 | −0.002 | −0.173 | 0.019*** | 4.691 |
Urbanity | −0.003 | −0.084 | −0.012 | −0.369 | 0.009*** | 3.851 |
Unemployed | 0.016 | 1.279 | 0.014 | 1.147 | 0.002 | 1.757 |
Financial problems | 0.106** | 3.746 | 0.097** | 3.199 | 0.01** | 2.833 |
Occupational category | ||||||
White collar | 0.012 | 0.685 | 0.011 | 0.587 | 0.002 | 0.948 |
Skilled blue collar | 0.073*** | 4.116 | 0.07*** | 4.069 | 0.002 | 1.086 |
Unskilled blue collar | 0.028 | 1.652 | 0.024 | 1.387 | 0.004 | 1.715 |
No occupation | 0.03 | 1.227 | 0.02 | 0.92 | 0.011 | 1.935 |
Immigrants at work | ||||||
Several immigrant colleagues | −0.094*** | −5.919 | −0.099*** | −6.043 | 0.005* | 2.744 |
Few immigrant colleagues | −0.058** | −3.166 | −0.06** | −3.285 | 0.002 | 1.04 |
Country-level variables | ||||||
Real size | −0.055 | −0.92 | −0.104 | −1.792 | 0.049*** | 5.04 |
. | Total . | Direct . | Indirect . | |||
---|---|---|---|---|---|---|
. | β . | t-ratio . | β . | t-ratio . | β . | t-ratio . |
Individual-level variables | ||||||
Education | −0.254*** | −9.075 | −0.226*** | −8.298 | −0.028*** | −5.021 |
Age | −0.003 | −0.186 | 0.007 | 0.392 | −0.01*** | −6.192 |
Female | 0.017 | 1.438 | −0.002 | −0.173 | 0.019*** | 4.691 |
Urbanity | −0.003 | −0.084 | −0.012 | −0.369 | 0.009*** | 3.851 |
Unemployed | 0.016 | 1.279 | 0.014 | 1.147 | 0.002 | 1.757 |
Financial problems | 0.106** | 3.746 | 0.097** | 3.199 | 0.01** | 2.833 |
Occupational category | ||||||
White collar | 0.012 | 0.685 | 0.011 | 0.587 | 0.002 | 0.948 |
Skilled blue collar | 0.073*** | 4.116 | 0.07*** | 4.069 | 0.002 | 1.086 |
Unskilled blue collar | 0.028 | 1.652 | 0.024 | 1.387 | 0.004 | 1.715 |
No occupation | 0.03 | 1.227 | 0.02 | 0.92 | 0.011 | 1.935 |
Immigrants at work | ||||||
Several immigrant colleagues | −0.094*** | −5.919 | −0.099*** | −6.043 | 0.005* | 2.744 |
Few immigrant colleagues | −0.058** | −3.166 | −0.06** | −3.285 | 0.002 | 1.04 |
Country-level variables | ||||||
Real size | −0.055 | −0.92 | −0.104 | −1.792 | 0.049*** | 5.04 |
Note: *P<0.05; **P<0.01; ***P<0.001 (two-tailed).
Education is by far the most important predictor of prejudice in our model.13 Our fourth research question concerned effect of education on prejudice that is mediated by perceptions of minority size. From Table 3 we can see that there is statistically significant indirect effect of education on prejudice that is mediated by size perceptions. Both the direct and indirect effects of education are negative and statistically significant, meaning that a part of educational effect on prejudice is due to more correct size perceptions of educated individuals. This finding shows that, to some degree, education reduces prejudice simply by increasing factual knowledge. The proportion of the total educational effect that is mediated by size perceptions is not very large, about 10 percent. However, we are only using one measure of relevant factual knowledge in our model, and it is possible that the impact of factual knowledge would have been stronger if we were able to measure respondents’ knowledge of other relevant facts about immigrants and immigration.
Apart from education, only Financial problems and Several minority colleagues have both a direct and indirect effect on prejudice. As mentioned previously, the real size of an immigrant population has a strong positive effect on the perceived size, and thus a highly significant positive indirect effect. Nevertheless, the total effect remains non-significant and negative.
Summary and discussion
We found that respondents perceived the size of the immigrant population to be larger than it really is in all 17 countries in our sample. The overestimates of the size of the immigrant population are statistically significant in all of the countries, are generally large and sometimes extreme.
Although the misperceptions are pervasive, we found a rather high positive correlation between the real size of the immigrant population and perceptions of this size. Thus, individual perceptions do not seem to be totally detached from the factual situation. This is confirmed by our multivariate analysis: by far the most important predictor of perceptions of size is the real size of the immigrant population in the respondent's country. Apart from real size, the education, age, sex, and levels of contact with immigrants and minorities influences how respondents perceive size.
Regarding education, highly educated individuals have more correct perceptions of immigrant size, but one might wonder why it is so. The information about size of immigrant populations is probably not emphasized in system of formal education, and even if it is, the relevant numbers are changing so fast that any information acquired during school years would be outdated rather quickly. A possible explanation for this seeming paradox is that information spread through mass media might be of importance. Relatively fast-changing and politically relevant information is obviously most efficiently spread to general population via news media. Although all strata of population in western countries have access to news media, it is known that better educated individuals are more capable of acquiring and retaining information from news media. This is also true in cases where information that is disseminated is not a part of school curricula (Price and Zaller 1993). Thus, it is possible that educated individuals are more able to obtain information about size of immigrant populations from their countries’ news media.
Perceptions of size also have consequences for attitudes toward immigrants: the larger the perceived size, the more negative the attitudes. In our multivariate model, the perceived size has the second strongest effect on anti-immigrant attitudes (i.e., prejudice), surpassed only by the effect of education. Interestingly, the real size of an immigrant population does not have an effect on prejudice in our model. While the effect of perceived size is easily explained in terms of increased threat perceptions (see, e.g., Quillian 1995), the association between real size and prejudice is a more complex one. Different postulations from conflict theory all predict increase in prejudice with the increased real size of minority population. Contact theory, however, predicts the opposite: an increase in the size of the minority should bring about a decrease in prejudice, provided that the level of contact between minority and majority groups increases with the increase in minority size. We do find evidence that increased contact with immigrants at work decreases prejudice, but this does not translate into the negative effect of immigrant size at the country level. Due to well-known problems of residential segregation and high unemployment levels among immigrants (e.g., Murdie and Borgegård 1998; van Tubergen et al. 2004), the assumption that increased immigrant size will in itself lead to increased contact with natives has to be questioned as being too simple. This is particularly so with regard to contact which takes place under ‘optimal contact conditions’ (i.e., equal status contact in the pursuit of common goals). Thus, our results remain inconclusive with regard to the effect of immigrant size in a country on anti-immigrant prejudice.
Given the ever growing immigrant population in Europe, the effect of the immigrant size on prejudice is probably one of the most important research topics today. However, the present empirical research based on the cross-country European surveys does not seem to provide any conclusive answer.14 Since the number of countries that can meaningfully be compared is inherently low, other research strategies will have to be developed in order to provide more clear-cut conclusions. Longitudinal studies (preferably panel studies) would be obvious candidates. The additional advantage of panel studies is that one could answer a question about possible presence of effect of prejudice on size perceptions.15 Another possibility is to focus more on studies on subnational level that analyse associations between minority size and prejudice in smaller geographical units such as regions and municipalities. In such cases, the advantage is that it is less problematic to obtain a higher number of observations at the aggregate level, but the disadvantage is that country-specific factors might make it is more difficult to generalise the results of one-country studies to other national contexts. For example, the two German studies presented earlier (Semyonov et al. 2004; Wagner et al. 2006) produce somewhat different results regarding minority size and prejudice: the former study does not find any statistically significant effect of minority size, while the latter actually finds a negative effect (i.e., less prejudice in areas with large immigrant populations, presumably due to better contact opportunities). The problem with both these studies is that they include both East and West German regions. It is known that East Germans are more prejudiced, in spite of comparatively low numbers of immigrants (Wagner et al. 2003), and it is also known that East Europeans generally tend to be more prejudiced than West Europeans (Strabac and Listhaug 2006). It is therefore possible that the methodologically very sophisticated study of Wagner et al. (2006) simply finds that East Germans are more prejudiced, and is therefore of little use outside of the German context.
Finally, our study shows that educated individuals are less prejudiced, and that this is partly due to these individuals' more correct perceptions of minority size. In other words, a part of the well-known effect of education on reduction of prejudice is simply a consequence of better factual knowledge. Interestingly, this effect is probably not a result of relevant information about minority size being provided as a part of school curriculum, but a result of educated individuals' better ability to obtain relevant information from their social surroundings. Thus, in order to reduce anti-immigrant prejudice ones should both provide correct information about immigrants in schools and news media, and work on increasing level of education in the country since education increases individuals' ability to absorb relevant information about immigrant populations.
Footnotes
Although, as we shall see, not necessarily information about size of minority populations.
The dataset was made available by the Norwegian Social Science Data Services.
The database is available at http://www.oecd.org/dataoecd/18/23/34792376.xls. For 13 out of the 17 countries included, the data are taken from a population census or from the population registers. German data are taken from the microcensus, Belgian data from the General Socio-Economic Survey, Finnish data from population statistics, and Dutch data from the Labour Force Survey.
Our main focus is on anti-immigrant prejudice of members of majority ethnic groups in the countries we analyse, and we therefore also remove non-immigrant individuals declaring themselves as belonging to an ethnic minority group from analyses. However, this has virtually no influence on our analyses, as the group in question is very small (342 individuals, about 1 percent of the total sample).
An accessible presentation of differences between multilevel models and clustering-corrected standard errors can be found in Primo et al. (2007).
There are several ways to measure the size of immigrant population in the country. We had to opt for percentage of foreign-born individuals since that is definition that was used in ESS survey question that is basis for our measure of perceived size. However, it is possible that respondents think of so-called ‘visible minorities’ when asked about immigrant populations, disregarding of clearly stated survey question and including the descendents of non-European immigrants. Thus, Silke Schneider is probably right when stating that improvement in measures of immigrant size should ideally be based on parents’ country of birth, so that also ‘second-generation immigrants’ might be included (Schneider 2008: 63). In other words, overperceptions of sizes of immigrant populations in countries that have large numbers of ‘second generation’ of non-European descent are perhaps partly caused by respondents’ inclusion of these in their immigrant size estimates.
We calculated two measures for the misperceptions. The first one expresses misperceptions simply as the difference between the perceived and the real size (i.e., in percentage points). The second expresses misperceptions as a proportion of the real size (in percent).
The UN estimates for 2002 show that 12.3 percent of the population in the USA was foreign-born (Borjas and Crisp 2005: 1).
None of the covariances among exogenous variables is assumed to be zero.
In multilevel regression analysis, the usual recommendation is to have at least 10 level-2 observations per level-2 predictor (Bryk and Raudenbush 1992).
It should be noted that this effect may not solely be a result of better factual knowledge of highly educated individuals. Other possible explanations include lower innumeracy among highly educated, lower overestimates of immigrant size due to prejudice, and lower proportion of immigrants in richer neighbourhoods (in which the highly educated tend to live).
More correct specification of the model would include a causal path from ‘real size’ to measure(s) of contact. We were able to estimate one such model, with dummy variable ‘several immigrant colleagues’ as measure of contact, and it turned out that real size has a positive effect on contact (β=0.183; P=0.013), but the main results of analysis otherwise remained essentially the same. A problematic aspect with this model is low value of Comparative Fit Index (CFI=0.345).
To some degree, the strong effect of education might also be a result of educated individuals' reluctance to admit anti-immigrant prejudice.
Nevertheless, one could argue that strong effects of immigrant size on prejudice would show, even with modest sample sizes on country-level. Therefore, if there are any effects of size on prejudice, they are probably quite weak.
One should keep in mind that the presence of this effect can not be disregarded, and that the effects of perceived size on prejudice estimated in our models might be somewhat upward biased.
References
Zan Strabac is associate professor of quantitative research methods and statistics at Trondheim Business School, University Sør-Trøndelag College, Trondheim, Norway.