What Determines Religious and Racial Prejudice in Europe? The Effects of Religiosity and Trust

Amendments to the Treaty on European Union and to the Treaty Establishing the European Community, Article 1, Preamble, http://​eur-lex.​europa.​eu/​legal-content/​EN/​TXT/​?​uri=​uriserv:​OJ.​C_​.​2007.​306.​01.​0001.​01.​ENG#a-001.

The lists of these countries are taken from http://​europa.​eu/​about-eu/​countries/​.

We realize that this may not be the best option. However, since there are no separate questions about accepting Protestant, Orthodox, and Catholic neighbors, we are forced to collapse the categories into one. We actually calculated the proportion of people from those three religions who said they don’t want “Christian” neighbors. The numbers were very small, which indicates that there is not much intolerance among those three religious groups in reference to neighbors.

We dropped the answer category “I don’t really know what to think” for doctrinal belief.

When analyzing clustered data using OLS regression, the total regression coefficient is actually a weighted combination of the within- and between-cluster regression coefficients (Raudenbush and Bryk 2002, p. 137). In multi-level modeling, when the models include the same predictors at different levels, it is important to use group mean centering of these variables [i.e., centering within cluster (CWC)]. Specifically, under CWC, the level 1 predictors are centered around the group mean of the cluster to which each case belongs (i.e., individual score minus the mean corresponding nation level score) (Enders and Tofighi 2007). Since the models include generalized trust and immigrant status at both levels, we group centered both of these variables. In addition, when testing the interaction effects that involve generalized trust and immigrant status, we also used the same group centered variables.

http://​epp.​eurostat.​ec.​europa.​eu/​portal/​page/​portal/​eurostat/​home/​. The macro data for Northern Cyprus is obtained from http://​www.​devplan.​org/​ since EUROSTAT does not have this information.

One disadvantage of this model is that it assumes the effect of covariates to be the same between countries. Assuming different effects of covariates requires the use of random slope MLR, which is beyond the scope of this paper.

As requested by an anonymous reviewer, we tested the correlations between all variables included in the final model (Model 6 in Tables 3, 4) for both religious and racial prejudice. We found some evidence to indicate the existence of multicollinearity only for religious prejudice. Thus, we also reran Model 6 in Tables 3 and 4 by excluding the immigrant status at the individual level, as well as all the religiosity measures except the religious denomination dummies. The results did not change, so we decided to keep all these variables in our final model.

We thank an anonymous reviewer for suggesting this exercise.