Explaining the standard errors of corruption perception indices

This paper examines the standard errors of two popular indices of corruption perceptions: the Worldwide Governance Indicators’ Control of Corruption (WGI-CC) and Transparency International's Corruption Perception Index (TI-CPI). The standard errors of these indexes stem from the degree of variation across the sources upon which these two aggregate indices are based. In general, standard errors are not associated with country characteristics; this supports the common assumption that differences across surveys are random. There are two exceptions, however. They involve the degree of media freedom in a country and the country's past corruption scores, possibly indicating the use of cognitive heuristics by the assessors who do the ratings. No evidence exists that more diverse countries have greater variation across corruption scores. In comparing the two aggregate measures, we find that the standard errors for TI-CPI are associated with country characteristics in fewer cases than are those for WGI-CC. Finally, our findings raise concerns about the applicability of the WGI-CC's use of the unobserved components model for extracting signals from noise.