Over the last decades, the rise of Social Capital as a central field of research in Social Science has contributed to an increasingly controversial and in-depth debate about trust. In particular, the identification of valid predictors for trust has been widely discussed, entailing various theories and outcomes. In the context of this course, this paper focuses on one specific predictor for trust, which has been generally accepted, namely the degree of diversity. As we live in an era of globalization, the topic of migration has become increasing relevant to the discussion of diversity (Castles and Miller, 2003). Many scholars have applied this relationship of diversity and trust, yet by using very diverse definitions of what they understand by diversity. For example, Putnam (2007) explores effects of a transition to a multicultural society based on categorizing people according to ethnicity as the diversifying feature. Delhey & Newton (2006), on the other hand, identify religious tradition as a central separating element when predicting trust.
This paper seeks at filling this gap of lacking conceptual clarification in terms of measuring diversity. We aim at answering the question whether or not the diversity measure has a significant impact on trust as the dependent measurement variable, and if so, which diversity variable matters most. For this research, we combine three papers related to the research question. Alesina et al. (2003) have presented a more comprehensive overview of diversity by introducing three different measures (ethnic, linguistic, and religious). Secondly, Fearon (2003) contributed to the topic through the refinement of the ethnic diversity measures, and the development of two indicators, ethnic and cultural fractionalization. Furthermore, he divided his large sample of countries into six world regions, a typology which we will also use in our study in order to identify possible peculiarities according to a certain region.
We will combine the three measures by Alesina et al. (2003) and the two by Fearon (2003) and correlate those five independent variables with the dependent variable “trust” on a country level, whose scores are taken from Bjørnskov (2008). The stronger the correlation, the more relevant we interpret the respective independent variable for trust. Moreover, we introduce viable controlling variables in order to eliminate external influences. After the already described screening for regional specialties, we finally execute a linear regression which would further explore the relationships between our variables of interest. We will argue on the basis of our findings that the diversity measure matters, in particular that ethnic fractionalization has the strongest impact on generalized social trust. However, after the inclusion of controlling variables, none of the independent variables appeared on a significant level – an outcome, which will be further discussed in our paper.
II. Methodology, Data Sources and Variables
In this paper we analyze the data from the papers by Alesina et al. (2003) and Fearon (2003) with the objective to re-examine the effects of diversity on trust, as measured by Bjørnskov (2008). We start by combining the different measures of diversity and the trust measure into one common dataset. We extract our first three variables - ethnic fractionalization, linguistic fractionalization, and religious fractionalization from the comprehensive overview of diversity by Alesina et al. (2003). They are measured through the probability that two randomly picked individuals of the population belong to a different group. A maximum value of 1 is reached when each person belongs to a different group as it can be seen from the following formula:
Abbildung in dieser Leseprobe nicht enthalten
[Abbildung in dieser Leseprobe nicht enthalten] - the share of group i (i =1,…, N) in a country j (Alesina et al., 2003, p.159)
The variable linguistic heterogeneity identifies linguistic groups on the basis of languages spoken as “mother tongues” within a country and the religious heterogeneity variable identifies religious groups within a country. The data for both is taken from Encyclopedia Britannica. For the variable ethnic heterogeneity, which identifies ethnic groups within a county incorporating also racial and language characteristics, the authors compare and combine data from Encyclopedia Britannica, CIA Factbook and Minority Rights Group International in order to achieve higher reliability (Alesina et al., 2003, p. 158-160)
The next two measures of diversity are taken from Fearon (2003). The ethnic groups are classified according to Fearon’s prototypical ethnic group using the following criteria:
1. Membership in the group is reckoned primarily by decent of both members and non-members
2. Members are conscious of group membership and view it as normatively and psychologically important to them
3. Members share some distinguishing cultural features such as common language, religion and culture
4. These cultural features are held to be valuable by a large majority of members of the group
5. The group has a “homeland” or at least remembers one
6. The group has a shared and collectively representative history as a group. Further this history is not wholly manufactured, but has basis in fact.
7. The group is potentially “stand alone” in a conceptual sense – that is it is not a cast or a cast-like group.
Source: Fearon, 2003, p.197
Fearon’s cultural fractionalization variable measures cultural resemblance between groups. The resemblance factor rij is an increasing function of the number of shared classification between languages i and j. rij is equal to 1 when one two groups speak exactly the same language and 0 when two groups’ languages come from completely different families: Cultural Fractionalization = 1 – Cultural Resemblance. The data were carefully chosen from sources – CIA’s World Factbook, Encyclopedia Britannica, the Library of Congress Country Study, and the Minorities at Risk data set (Fearon, 2003, p. 202).
The last variable – trust - Bjørnskov measured “in what has become the standard way, by taking the percentage of a population that answers yes to the [World Values Survey] question ‘In general, do you think that most people can be trusted” (Bjørnskov, 2008, p.275).
The data for the six variables includes different numbers of countries from different periods. Since the variable trust will be our dependent variable we stick to the countries from Bjørnskov’s paper but we delete Serbia and Montenegro, because no consistent data for them is available in the other sources. It should also be noted that we modify the data for Germany from Fearon’s paper in the following way: we weigh the scores for the German Democratic Republic and the Federal Republic of Germany in proportion 1:2 on the basis of the current population distribution in Eastern and Western Germany. There is missing data on the linguistic fractionalization in El Salvador, on Alesina et al.’s (2003) ethnic fractionalization for Puerto Rico, and on the two variables from Fearon for Hong Kong, Iceland, Malta, and Puerto Rico, so these cases are automatically excluded pairwise by SPSS from the analysis. Furthermore, we divide this final set of countries into six world regions, using the same typology as Fearon – Western Europe and Japan, Eastern Europe and the Former Soviet Union, Asia, North African and the Middle East, SubSaharan Africa, and Latin America and the Caribbean (Fearon, 2003, p. 215 ff.). This is done in order to identify possible peculiarities according to a certain region. After we have the final data set we identify the examination technique: First we use the Pearson Product Moment Correlation to check the relationship between our independent variables and trust – we correlate the five diversity measures (our independent variables) with the dependent variable “trust” on a country level, without controlling. The stronger the correlation, the more relevant the respective independent variable will be for trust. The same test will be run after splitting by regions. Next we will introduce controlling variables in order to eliminate external influences. We will use the Partial Correlation to check the relationship between our independent variables and trust when controlling step by step for:
1. Wealth: Measured with GDP/capita- (CIA Factbook)
2. Income inequality: Measured with the GINI coefficient (CIA Factbook)
3. Rule of law: Taken from the World Bank Good Governance Indicators
The last step of our analysis consists of a regression with the two significant ethnic fractionalization measures and the control variables to investigate which of them has the greatest influence on trust.
Bivariate Correlations: Aggregate Level
In order to investigate the relationship between ethnic diversity we conduct correlation tests with all of the five different diversity measures on both the aggregate and regional level. If some kind of association exists it should appear at least with some of the diversity measures. The results of the test are represented in Table 1. Contrary to popular belief, only two of the diversity measures are significantly correlated with trust. According to the results, there exists a significant negative relationship between trust and Alesina et al.’s (2003) ethnic fractionalization measure (r=-0.275, p=0.006), meaning that the more ethnically fractionalized the society is, the less trusting its members. The same pattern could be observed with Fearon’s (2003) ethnic fractionalization measure although the correlation is weaker and significant only at the 0.05 significance level (r=-0.220, p=0.031). Although the correlations between trust and the ethnic fractionalization measures are significant, they are not that strong. This fact combined with the non-significant relations with the other measures brings the destructive effect of diversity on trust into question. In order to clarify the roots of the association between the two ethnic diversity measures and trust, we decided to investigate the character of these associations across the different world regions.
Bivariate Correlations: Regional level
From Table2 and Table3 one could see that there exist clear differences among the world regions in terms of GDP and ethnic fractionalization as measured by Alesina et al. (2003). Additionally, the one-way ANOVA showed that the six world regions differ significantly in terms of ethnic diversity with F(5)=7.891, p<0.0005. These differences could be more clearly seen from Graph 1 which illustrates the diversity and trust levels among the six world regions. According to the graph, the six world regions exhibit quite different trust and diversity patterns, with the African countries representing an especially interesting case with their high levels of diversity and low levels of trust. From the graphs it could also be concluded that a clear linear association between trust and ethnic diversity is simply not observable at the regional level. However, to further investigate this observation, we conduct simple correlation tests between the two significant ethnic fractionalization measures from the first part of our analysis (see Table1), and trust in each of the world regions. The results of this analysis are presented in Table5. None of the significant associations that previously appeared are present in any of the world regions. Moreover, the correlations change their directions from one world region to another, which renders the alleged trust-diversity association even more dubious and questionable. One possible reason for the weak results on regional level could the poor quality of the data of the underdeveloped countries together with the small regional samples. Another explanation could be that there exist some other variables that influence the levels of trust within a society even stronger than ethnic diversity, and these variables have already shaped the differences of the separate world regions. In order to account for this differences we conduct partial correlation tests controlling fro GDP per capita, GINI coefficient and rule of law to see if the trust-diversity association is still significant.