Abstract
Drawing on data from the World Values Survey (1995–2014), and using a multilevel modeling technique, this article explores the relationship between social inequalities and religiosity for a sample of 223,016 respondents nested in 85 countries. Mixed-effects regression models find that income inequality is a stronger determinant of individuals’ religiosity in developing (lower middle-income) countries compared to developed (higher middle-income and high-income) nations. Alternatively, healthcare inequality, operationalized as the out-of-pocket share of a country’s total health-related expenditures, is a stronger predictor of individual-level religiosity in high-income countries, relative to both upper and lower middle-income nations. These findings broaden our understanding of the global religious dynamics through discovering a cross-national heterogeneity in paths through which the process of secularization works in developed and developing nations.
Notes
Acknowledgments
The author would like to acknowledge Dr. Rob Clark and Dr. Martin Piotrowski for their helpful comments and thoughts on the earlier versions of this paper.
Notes
6 A list of countries and their corresponding numbers of observations is provided in Appendix A.
7 For more information and a detailed discussion of Welzel’s indices, refer to the online appendix of Freedom Rising: Human Empowerment and the Quest for Emancipation available at: www.cambridge.org/cl/download_file/473755/
8 To check the robustness of the religiosity index, I used an Item Response Theory (IRT) test. IRT is a technique built upon CFA specifically developed to evaluate the quality and relevance of multidimensional measures (Hekmatpour and Burns Citation2019; Kean and Reilly Citation2014). This test evaluates the quality of an index by estimating the correlation between a predicted latent factor loading on the observed items and the measure created by the researcher (Harvey and Hammer Citation1999). In the analyses, this correlation is strong (r = 0.965) and statistically significant (p < .001), which supports the relevance of the religiosity index used in this research.
9 A complete list of countries in these three categories can be found in Appendix B.
10 I tried to control for more country-level variables, such as level of human capital, the share of the population that has access to the internet, the share of labor force active in agriculture, and level of access to improved water resources. However, these variables turned out to be highly correlated with GDP per capita, thus causing a multicollinearity problem in statistical models. Therefore, I decided not to include these country-level indicators of development in the models.
11 I use listwise deletion which excludes observations with missing values on all variables from the analyses.