PLAN DE CUENTAS

Interest from academic economists in investigating the causes of corruption has followed largely from the inuential work of Mauro (1995). In that paper, the author presents evidence regarding the negative eects of corruption on economic performance. As we have already noted in Chapter 2, while the number of cross- country comparative empirical studies on the determinants of corruption has in- creased, there appears to be little consensus on the eect of any variable on cor- ruption apart from economic development and democracy. To give an example: the variables that have received most attention as determinants of corruption in the literature are British colonial heritage, uninterrupted democracy, protestant religion, electoral rule, and scal decentralisation [see Ades and Di Tella (1999),

La Porta et al. (1997, 1999), Treisman (2000), Persson et al. (2003), and Ad- sera et al. (2003)]. Yet, these studies have obtained mixed ndings on the same explanatory variables, possibly due to the use of dierent corruption indicators, dierent samples of countries, and perhaps most importantly the use of a diverse set of conditioning variables within their empirical specications. Despite this sensitivity of the results there remains within the literature little systematic re- search on the robustness analysis of the determinants of corruption. Some recent exceptions are the global sensitivity analysis by Serra (2006) and Seldadyo and de Haan (2005, 2006). They use Extreme Bounds Analysis (EBA) as modied by Levine and Renelt (1992) and other versions [Sala-i Martin (1997)], to perform an analysis of the sensitivity of the regression estimates to changes in the pool of control variables. Our present study focuses on a variable used in those papers, press freedom, uses time series-cross section data and suggests some further modi- cations to EBA. We also examine the relationship between particular restrictions to press freedom and corruption.

The study of press freedom as a determinant of corruption has, compared to the prominent variables mentioned so far, been largely absent in the literature4_{. At-}

tempts to introduce the topic have come from a group of papers whose main focus has been placed exclusively on press freedom, rather seeing this as one determi- nant of corruption amongst many others. Brunetti and Weder (2003) test the hypothesis that a free press should a priori be associated with lower corruption. The authors use a press freedom measure compiled by Freedom House (who also publishes the popular indexes of political rights and civil liberties). The index ranks countries according to a 0-100 scale, with low values meaning a high de- gree of press freedom. Using this and other alternative measures for both press freedom and corruption, they nd that the empirical evidence suggests a strong negative relation. Their result is robust to controlling for alternative specications and econometric methods. They conclude that in countries where the media is

4_{Adsera et al. (2003) used a proxy for the diusion of newspapers and found a signicant and}

large coecient. As others and we have discussed, this proxy does not appropriately reect the freedom that journalists and reporters enjoy.

reasonably free from any kind of restriction on their activities corruption levels are likely to be low. Although the authors acknowledge that the problem of causality may be dealt with instrumental variables and using some panel-data techniques, the choice of the instrumental variable for which they obtain their main ndings is somewhat debatable5_.

Similarly, Ahrend (2002) examines the relationship between the variables from a wider perspective. His objective is to study the relationship between human capital, press freedom and corruption. He notes that a high degree of press freedom acts as a channel through which education decreases corruption. Additionally, the author nds evidence suggesting that high corruption levels are associated with low levels of press freedom. The causal direction, according to his work, runs from a freer press to lower corruption. Chowdhury (2004) presents a concise treatment of the topic. The objective is similar to Brunetti and Weder (2003) but also incorporates the eects of democracy on corruption. In his view, the media's role as an informative device and the standing of democracy as a punishing mechanism should both help towards limiting corruption. The empirical ndings of the paper support this conclusion: both press freedom and democracy are powerful and signicant controls on corruption and this result is robust to dierent settings. The author remains cautious regarding the direction of causality. Finally, Lederman et al. (2005) examine the relationship between several political institutions and corruption. They nd evidence of an association between freedom of the press and corruption. However, the coecient on press freedom becomes insignicant when they include a control for economic development in the corruption regression. While all these studies reach the same conclusion that press freedom is bad for corruption, they use an aggregate measure of press freedom. Additionally, the variables included in the base specication are not always those suggested by the

5_{They use an index controlling for political rights. This variable, although highly correlated}

with press freedom, is also correlated with corruption measures as we have described earlier, which violates one of the conditions required for a variable to be a good instrument. The authors acknowledge this may be a problem but they argue that it is reasonable to assume that it is a good and valid instrument. In other section of their paper, they use dierent instruments and their main ndings are conrmed.

empirical literature as robust predictors6_{. Finally, most of the empirical literature}

does not include strong tests for robustness and those which do found mixed evidence. For example, Serra (2006) and Seldadyo and de Haan (2006) have used proxies for press freedom in their studies using sensitivity analysis. They nd that press freedom is not robust according to the EBA methodology, implying that the relationship between this variable and corruption is sensitive to changes in the specication.

In the present chapter, in addition to testing for the robust relationship between the aggregate press freedom and corruption, we use previously unexplored data on dierent forms of restrictions on press freedom and to test the robustness of their individual relationships with corruption. Further, from a broader perspective, our chapter may be seen as an extension of the literature on Extreme Bounds Anal- ysis (EBA) that has been originally proposed by Leamer (1983, 1985) and made popular by Levine and Renelt (1992) in the context of cross-country growth re- gressions. In order to avoid some well-known criticisms of standard EBA being too restrictive for a potentially important variable to pass as robust [Sala-i Martin (1997)], we carefully screen the regression models for potential problems of simi- larity, collinearity and t that may help to explain why a variable is not robust. We describe these modications later in this chapter.