Corruption and the Extractive Industries Transparency Initiative

Abstract

The Extractive Industries Transparency Initiative (EITI) has received much attention as a scheme that can help reduce corruption in mineral-rich developing economies. To our knowledge, this paper provides the first empirical attempt (using panel data) to explore how EITI membership links to changes in corruption levels. We also examine whether the different stages in EITI implementation (initial commitment, candidature, full compliance) influence the pace of changes in corruption. We find that EITI membership offers, on the whole, a shielding mechanism against the general tendency of mineral-rich countries to experience increases in corruption over time.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. When we refer to mineral resources, we refer to both oil and non-petroleum minerals.

2. The implementation process and rules are described in detail at the EITI Standard: see https://eiti.org/document/standard.

3. Some candidate countries are more successful than others in reaching full compliance – for example, Norway achieved compliance within approximately two years, while Afghanistan has been a candidate country for almost 5.5 years.

4. We carried out a number of diagnostics tests prior to proceeding with our empirical model. We carried out the Ramsey test for misspecification (by examining whether non-linear combinations of the fitted values help explain the actual changes in corruption) and in all cases we rejected the hypothesis that there are omitted variables. We created several non-linear combinations of fitted values (starting from the simple quadratic term of the fitted values and progressively adding higher-order terms, up to the power of eight), and then check if these were jointly statistically different to zero. We also carried out the Breusch-Pagan test for heteroscedasticity and in all cases we found that the error term is homoscedastic. In any case, we cluster our standard errors at the country level to allow for arbitrary country-specific correlation of errors.

5. We do not include the EITI dummy as a separate explanatory variable due to its high correlation with the interaction term (0.71, 0.75 and 0.85 for the case of commitment, candidature and compliance respectively) to avoid multicollinearity problems. This is discussed in several empirical papers in the literature, where it is not uncommon to drop some of the highly correlated component terms (for examples, see Etang, Fielding, & Knowles, 2011; Huang, 2008). Intuitively, one would also not expect EITI participation to have a direct effect on corruption, that is, beyond the corruption-reducing effect that it might have for mineral rich member countries.

6. Online Appendix 1 lists countries in the sample (for the richer specification of Table 1). Online Appendix 2 provides a correlation matrix for all variables appearing in the analysis and Online Appendix 3 lists all variable descriptions and data sources. Descriptive statistics are presented in Online Appendix 4.

7. In 2012, Transparency International changed the way of measuring its Corruption Perception Index – the new index is hence not comparable to its pre-2012 equivalent. Replicating our specifications for the 2012–2014 period (when data are again comparable) provides very similar results.

8. As a result of such limited within-cross-section variance, the Hausman test statistic also has insufficient power to select between fixed-effects and random-effects estimations and is, hence, inappropriate (for a discussion, see Baltagi, 2011, p. 321; Christen & Gatignon, 2011; Clark & Linzer, 2015).

9. The EITI dummy takes a value of 1 both for the year during which the country reaches the corresponding stage in EITI implementation (commitment, candidature, compliance), as well as for all consecutive years.

10. The mineral dependence proxy is unlikely to be endogenous to changes in corruption – the level of mineral dependence is the result of combination of geography and long-term investment, while changes in corruption are yearly (that is, short-term). We run a series of Granger causality tests that indeed showed that the causality runs from mineral resources towards changes in corruption. Another possible endogeneity concern relates to whether it is not only EITI participation that influences changes in corruption, but whether the decision behind EITI participation is itself endogenous and dependent on the level of corruption. We run a series of Probit regressions where the EITI participation dummy is regressed on the level of corruption in the previous period, and/or other lagged variables (level of GDP per capita, democracy, and so forth). In none of these did we find the coefficient of corruption (or the corresponding average marginal effect) to be statistically significant.

11. We have also replicated the table for the case of non-petroleum minerals – results are in line with our earlier findings.

12. We have also replicated the specifications for the case of EITI compliant countries – the coefficients of both mineral and oil abundance, as well as those of the interaction terms, are all insignificant.

13. For Columns 19 and 21, the period of analysis extends to 2012 – for Column 20 to 2013.

14. The interaction terms remain statistically insignificant when results are replicated for the case of compliance.