Measuring world governance: revisiting the institutions hypothesis

See, for example, Neumayer (2002), Easterly and Levine (2003), Beck and Laeven (2006), Easterly (2007), Durlauf et al. (2008) for the similar use of the WGIs.

See, for example, Dollar and Kraay (2003), Rodrik et al. (2004), Bulte et al. (2005), Harms and Lutz (2006); Aidt et al. (2008), Brunnschweiler (2008), Bosker and Garretsen (2009) for the similar use of the WGIs.

Space restrictions preclude us to present the best-case scenario WGIs for each year, but we present the best-case scenario for each governance indicator. The results for the best-case scenarios for each year are available upon request from the author.

Similar applications employed by Pinar et al. (2013) to construct the best-case scenario of Human Development Index and Agliardi et al. (2012) to construct the riskiest sovereign risk index for the emerging countries. The current study differs from the above mentioned ones in two ways. First, both Pinar et al. (2013) and Agliardi et al. (2012) examine the absolute levels of human development and sovereign risk, respectively. However, in the current paper, we examine the relative levels of governance indicators. In this case, we derive the least volatile governance indices and therefore these indices will be less subject to measurement error. Second, the current study examines the effect of the use of the proposed indices in the relevant literature as institutional quality proxies and provides further insights on how measurement could be an important issue.

As each governance indicator is bounded between \(-\)2.5 and +2.5, higher measured governance levels for more countries suggest a distribution that is negatively skewed and therefore having lower variability across countries and over time.

Each aggregate governance indicator is highly correlated with the other aggregate governance indicators at a given year. Moreover, each governance indicator outcomes at a given year is highly correlated with the following year’s outcomes of that indicator. The correlation coefficients, both the simple and the Spearman-rank ones, are very high (above 0.9) and very significant.

We have defined above \(\varvec{\lambda }\) and \(\varvec{\tau }\) to be different weighting vectors that are associated with different indices. In the discussion that follows, we use \(\varvec{\lambda }\) and \(\varvec{\tau }\) interchangeably with the index that they represent.

Test statistics are obtained for different governance levels considering all possible weight combinations which require mixed integer and linear programming for the first- and second-order SDE test statistics. See Sect. 4 of the ST for the mathematical formulation details.

We thank an anonymous referee for suggesting us to test for the first-stage best-case scenarios.

The next sections summarize the findings of the three studies when the best-case scenario governance indices are used as institutional proxies. However, detailed findings of the three studies are available upon request from the author.

To assess the robustness of the findings in the next sections, we also test whether the equally weighted governance indices are the best-case scenario indices or not for the full sample size which consists around 200 observations. We find that the equally weighted indices are not the best-case scenarios for all governance indices when the full sample size is used. Even though weights assigned to each year in the case of full sample size differ from the ones that are found with different sample sizes, the findings that are presented in the next sections do not change significantly when the best-case scenario governance indices for the full sample size are used as institutional proxies. Especially, the different results with the revisited papers still hold when the best-case scenario indices for the full sample size are used. Full sample size best-case scenario governance indices and detailed findings of three studies with those indices are available upon request from the author.

BG (2009) cited Rodrik et al. (2004) paper that the available language instruments for large sample do not pass over-identification test, whereas in the previous section, we find that the language instruments are valid for some of the best-case institutional quality indices. However, we continue with BG specification.