Corruption clubs: empirical evidence from kernel density estimates

Abstract

A common finding of many analytical models is the existence of multiple equilibria of corruption. Countries characterized by the same economic, social and cultural background do not necessarily experience the same levels of corruption. In this article, we use Kernel Density Estimation techniques to analyse the cross-country distribution of corruption. Particular emphasis will be given to the question whether the distribution of corruption shows more than one peak. We find that most of the estimated densities exhibit twin peaks. We also provide some evidence on the intra-distribution dynamics and the persistence of corruption. We find the group of countries classified within the two ‘clubs’ to be very stable. Corruption is a highly persistent phenomenon. Substantial changes in the economic, political and cultural environment of countries within the ‘corruption club’ are required before a significant decline of corruption is to be expected.

Notes

¹ Surveys of this literature include Bardhan (1997), Tanzi (1998) and more recently Jain (2001) and Aidt (2003).

² Andrianova (2001) extends this model by considering collective as well as individual reputation.

³ In a recent survey, Jain (2001) illustrates the complexity of the relationship between corruption and the effectiveness of a country's legal system by means of a simple example: ‘The level of corruption in a country with an ineffective legal system may begin to rise in response to, say, an external shock. The political elite may find the increased income from corruption irresistible. Once corrupted, the elite will attempt to reduce the effectiveness of the legal and juridical systems through manipulation of resource allocation and appointments to key positions. Reduced resources will make it difficult for the legal system to combat corruption, thus allowing corruption to spread even more’ (p. 72). This inter-relation is also stressed for Russia by Levin and Satarov (2000). In the context of transitional Russia, they propose that various forms of corruption are created due to an underdeveloped body of legislation that makes manifest itself in poor awareness of the law, ill-defined legislative procedures and inconsistencies among legislative acts. At the same time, Levin and Satarov stress that ‘corruption infiltrates courts’ (p. 123). Lawyers can use bribes as an effective tool for the defence of their clients. More systematic evidence for the inter-relationship between corruption and rule of law is provided in Herzfeld and Weiss (2003).

⁴ For more details see Lambsdorff (2002).

⁵ The methodological background of the Kernel Density Estimation is described in detail in Silverman (1986) and Wand and Jones (1995).

⁶ We perform different robustness exercises to show that results do not change substantially with different smoothing parameters. Despite the fact that Silverman's rule of thumb tends to over-smooth bimodal distributions (Silverman, 1986; p. 47), the results when using other methods are very similar.

⁷ The CPI shows a third mode around 7.4 which is, however, very weakly distinct.

⁸ The only cases where a unimodal distribution is not rejected is the Graft index in the last 2 years, the IMD index in 1991, and the WEF index in 2002. The corruption data from the CPI in 1997 as well as the IMD in 1990 and 1992 suggest more than two modes. One could argue that bimodality is a result of sample selection since the largest sample (Graft index 2002) shows a unimodal distribution. But running the bootstrap test for a smaller sample for the years 2000 and 2002 reveals the same result; the unimodality hypothesis is not rejected at a 5% significance level.

⁹ The following simple rule is used for allocating countries: i ∈ G_j if x_i  ∈ [c_j , c_{j +1}] for j = 1, 2 where c_j are the cut-points defined as the values of x at which the estimated density has a local minimum.

¹⁰ For details see Scott (1992) and Silverman (1986), Chapter 4.