The relationship between religion and corruption: are the proposed causal links empirically valid?

Abstract

There is a growing interest in understanding how religion affects corruption. Many empirical studies have suggested that countries with strong hierarchical religions (such as Islam, Catholicism and Orthodox Christianity) are more likely to suffer from corruption. These results are, however, controversial, largely due to the lack of empirical validity of the causal (or theoretical) links that explain such a relationship: obedience to authority, negative culture reinforcement, amoral familism and trust intermediation. Using the fourth wave World Values Survey (n = 87,988) of 64 countries, this study constructed a general estimation equation model to evaluate these four causal links after controlling for heterogeneity of individuals’ religious beliefs among the 64 countries. We did not find strong evidentiary support for the causal explanations.

Keywords:

• religion
• corruption
• trust
• religious commitment

Notes

1. CPI is calculated using multiple sources that differently assess the corruption level of the same country. The average standard deviation of the 2004 CPI was 0.81. The measurement error of CPI is not controlled in most empirical studies.

2. The trust model implicitly assumes that trust is a cause of corruption, not a result. However, the endogeneity between trust and corruption should be tested, as Uslaner (2004) suggested. This study tested endogeneity using a Hausman test.

3. Some observations of variables used in multiple regression are missing (around 15%); this might cause some bias. We used multiple imputation, a Markov chain Monte Carlo technique; findings did not change, while estimated coefficients varied.

4. The perception measure’s country sample sizes ranged from 153 (Israel) to 4585 (Turkey).

5. In the third-wave WVS survey, corruption was directly measured by the question, ‘How widespread do you think bribe taking and corruption is in this country?’ However, the question was not included in the fourth wave. According to Ko and Samajdar (2010), CPI, WVS and actual bribery levels of the International Victimization Survey are highly correlated.

6. The classification of religion varies depending on the research question. For instance, Paldam (2001, p. 393) classified religions into four groups: monotheistic (Judaism, Islam and Christianity), polytheistic (Hinduism), systems of belief (Confucianism) and tribal religions and atheism.

7. The Spearman rank correlation between social class (five-point scale) and income (10-point scale) is 0.42 (p-value < 0.01).

8. This approach is similar to the fixed-effect multilevel model. The GENMOD procedure of SAS 9.2 was used for the general estimation equation method.

9. Some readers might be interested in factors that are more important in explaining the obedience emphasis. When we checked the influence of independent variables on the choice of obedience using the Wald chi-square test, country (95%) and religion (3%) explained most chi-square statistics. This suggests that country-specific variation is far more critical than religion.

10. Represented as an odds ratio, the 11% point difference would be 0.59 or (24/76)/(35/65).