Abstract
In this paper, we employ, and test the reliability and validity, of multiple corruption related datasets such as the Corruption Perceptions Index of Transparency International, the Control of Corruption Index of the World Bank, the International Country Risk Guide of PRS, the World Value Survey, the Bribe Payers Index, and the International Crime Victimization Survey. We find evidence that the reliability of international corruption indexes has improved over the years. However, a number of shortcomings – such as the risk of selection bias, longitudinal sensitivity, and measurement errors that are inherent in composite indexes – are commonly observed. These problems are critical enough to affect the results of empirical analysis, as we illustrate using linear models. Therefore, we conclude that researchers should pay more attention towards minimizing the impact of measurement error through rigorous data screening and robustness tests using multiple data sources and methods. A thorough and systematic evaluation of the reliability and validity of international corruption indexes is essential to increase their usefulness.
Notes
1 Assistant Professor, National University of Singapore, Adjunct Staff, RAND Cooperation.
2 Ph.D. Candidate, National University of Singapore.
3 Lambsdorff, Johann Graf, “Measuring the dark side of human nature: The birth of the Corruption Perceptions Index” (http://www.icgg.org/corruption.cpi_childhooddays.html) accessed last on 1st March 2010.
4 http://www.transparency.org/policy_research/surveys_indices/cpi/previous_cpi/1999/reactions (last accessed 1st March 2010).
5 http://www.transparency.org/policy_research/surveys_indices/cpi/previous_cpi/1999/reactions (last accessed 1st March 2010).
6 When the Heritage Foundation uses CPI, it defines corruption as “dishonesty or decay”.
7 Please refer to Appendix 1 to see how empirical studies use international corruption indexes.
8 CitationYou and Khagram (2005) state that ratings based on perceptions and subjective judgments of experts can be imprecise and possibly biased. CitationMontinola and Jackman (2002) state that perceptions based measures cannot provide a count of the actual incidence of corruption, and can only indicate variations across countries in the overall climate of corruption.
9 BI published indices on 56 risk factors for 68 countries, for the period 1980–1983, and on 30 risk factors for a sample of 57 countries for the period 1971–1979. Economist Intelligent Unit took over the BI index in the end of the 1980s and Transparency International and World Bank use it to construct their own corruption indexes. Therefore, in this paper, we do not perform independent validity and reliability tests for BI.
10 African Development Bank's CPIA, Asian Development Bank, World Bank's CPIA, EIU, Freedom House, GI's Business Conditions and Risk Indicators, IMD, MIG, Political and Economic Risk Consultancy, and WEF's GCS.
11 Since higher CPI score implies lower level of corruption, the negative correlation implies that the low perception of corruption is related to low level of bribery.
12 also suggests that bribery is a very minor problem in clean countries where citizens have little experience with bribery.
13 The correlation coefficient is 0.62 (N = 18, p-value=.005) in year 2004, and 0.88 (n = 15, p-value <.0001) in year 2003.
14 The population and GDP per capita of 20 countries of Group 1 are not known because of lack of information. The average WBCC of these 20 countries is 0.53 (95% confidence intervals: 0.16–0.90), which is the similar to the other 10 countries whose population and GDP per capita data are available (mean: 0.74, confidence interval: 0.34–1.15).
15 We do not control for other variables since our goal is not to test hypotheses regarding causes or effects of corruption.
16 Since dependent variables have different scales, we cannot compare the absolute coefficient sizes but we can indirectly compare them using p-value.
17 According to Social Weather Stations’ national survey from 1987 to 2005, in Philippines, the rich consistently are less satisfied with the National Administration in Eradicating Graft and Corruption compared to the very poor income class (http://www.sws.org.ph/).
18 The classification scheme used here categorizes definitions according to their emphasis on the economic, institutional or moral aspects of corruption. This differs from the classification scheme widely used in the literature. The classical or the moral approach and the contemporary or revisionist or behavior-classifying approach have been identified in the literature as the two broad approaches to corruption (CitationCaiden & Caiden, 1977; CitationJohnston, 2001; CitationNye, 1967). The two schemes most widely used to classify contemporary definitions are the public office-public interest-public opinion scheme (CitationGardiner, 1993; CitationKurer, 2005) and the public office-public interest-market centred scheme (CitationCaiden and Caiden, 1977; CitationJohnston, 2001).