ABSTRACT
Explaining negative attitudes towards immigration in general and threat due to immigration, in particular, has been a major topic of study in recent decades. While intergroup contact has received considerable attention in explaining ethnic threat, group relative deprivation (GRD), that is, feelings that one’s group is unfairly deprived of desirable goods in comparison to relevant out-groups, has been largely ignored in cross-national research. Nevertheless, various smaller-scale studies have demonstrated that GRD can have a decisive impact on prejudice. In the current study, we examine the association between GRD and ethnic threat systematically across 20 European countries, thereby controlling for intergroup contact and value priorities. The 7th round of the European Social Survey (ESS) includes questions assessing respondents’ feelings of group deprivation compared to immigrants and offers for the first time an opportunity to contextualise the threat-inducing effect of GRD across Europe. A multilevel structural equation model (MLSEM) demonstrates a considerable association between GRD and ethnic threat both on the individual and country levels. The results indicate that GRD is not only an important mediating factor between social structural positions and perceived threat, but also fully mediates the relation between contextual economic indicators and ethnic threat.
Acknowledgements
The work of Peter Schmidt was supported by a fellowship of the Polish Foundation for basic research. Eldad Davidov would like to thank the University of Zurich Research Priority Program Social Networks for their support. The authors would like to thank Lisa Trierweilr for the English proof of the manuscript.
Disclosure statement
No potential conflict of interest was reported by the authors.
ORCID
Bart Meuleman http://orcid.org/0000-0002-0384-5995
Eldad Davidov http://orcid.org/0000-0002-3396-969X
Notes
1 See, for example, the joint statement of the Office for Democratic Institutions and Human Rights (ODIHR), the European Commission against Racism and Intolerance (ECRI) and the European Union Agency for Fundamental Rights (FRA) of 21 March 2009 (http://fra.europa.eu/sites/default/files/fra_uploads/355-evt-21March-jointstatement-09_en.pdf).
2 In line with Blumer (Citation1958), we define perceived threat as an individual perception that minority groups pose a threat to the in-group social position and the established social order.
3 Several studies combine realistic and symbolic threat perceptions into a single general factor of threat because it is not possible to distinguish between the two aspects (see, e.g. Hercowitz-Amir, Raijman, and Davidov Citation2017).
4 A similar argument could be made for income inequality (e.g. Wilkinson and Pickett Citation2007), but this is beyond the scope of this paper.
5 Of all individual characteristics, we only include GRD at the country level. GRD is the focus of this analysis, and we find it essential to include it on both levels of analysis. The reason that we do not include any further country-level predictors is that the limited sample at the country level forces us to formulate a parsimonious between-level model.
6 We excluded Israel from the analysis because of the distinct character of immigration and ethnic minorities in this country.
7 Confirmatory factor analyses suggested that the items included in the ESS immigration module were equivalent across countries, thus allowing a meaningful comparative analysis (Davidov, Cieciuch, and Schmidt Citation2018)
8 We would like to note that whereas the large meta-analysis of Smith et al. Citation2012 on GRD covered 29 countries in total, our data allowed investigating the effect of GRD in 20 countries in a single study with the same measures.
9 4,861 persons reporting no intergroup contact at all were obviously not asked to evaluate the quality of this contact. These persons have a missing value on the quality of contact variable. Since Full Information Maximum Likelihood (FIML; Schafer and Graham Citation2002) rather than listwise deletion was used, however, these observations are not removed from the analysis.
10 For the specifications of the Bayesian estimation, we followed the procedures described in van de Schoot et al. (Citation2014). All prior distributions were specified to be non-informative with the default N(0,∞) for factor loadings and intercepts and IG(-1,0) or IW(0,-3) for (co)variances. We assessed model convergence using the Gelman-Rubin criterion (Gelman et al. Citation2004) with 0.01 as the cut-off value (Hox, van de Schoot, and Matthijsse Citation2012). Furthermore, we requested two different chains of the Gibbs sampler and checked convergence visually by inspecting trace plots for all parameters. Since some between-level parameters displayed autocorrelation (i.e. parameter values for consecutive draws show similarity), we used a thinning factor of 50, and increased the number of effective draws to 10,000. The Kolmogorov-Smirnov test comparing the posterior distributions for the chains confirmed convergence for all parameters.
11 We use Bayesian estimation because of its good small-sample size performance, and not for a principled rejection of the practice of null-hypothesis significance testing. Therefore, we do provide p-values in the result section. These p-value represent the proportion of estimates over the iterative procedure that has a value smaller than 0 when the parameter is positive, or larger than 0 when the parameter is negative. The p-values are thus one-sided p-values.
12 Note that the Bayesian estimation procedure implemented in Mplus does not allow the inclusion of weights, so that we cannot use weights to correct for cross-national differences in sampling design (weight factor dweight). Re-estimating the model with the Maximum Likelihood estimator and the design weight does lead to very similar results, however.