The welfare costs of corruption

Abstract

Corruption has been shown to affect a variety of economic indicators, especially GDP per capita. However, as GDP is not a genuine indicator of welfare, it may reflect the welfare costs of corruption only in an incomplete way. This article uses self-rated subjective well-being as an empirical approximation to general welfare and shows that cross-national welfare – operationalized in this way – is affected by corruption not only indirectly through GDP, but also directly through nonmaterial factors. This article estimates the size of these effects as well as their monetary equivalent. The direct effect – not previously investigated in the corruption literature – is found to be substantially larger than the indirect effect.

Acknowledgements

I am grateful to Udo Ebert, Klaus W. Schüler, an anonymous referee and seminar participants at the University of Oldenburg for useful comments.

Notes

¹See www.icgg.org.

²An early study of the economics of happiness is Easterlin (1974). Later contributions examine the relationship between income distribution and self-rated happiness (Morawetz et al . 1977) and between unemployment and happiness (Clark and Oswald, 1994; Winkelmann and Winkelmann, 1998). Di Tella et al . (2001) use data from happiness surveys to estimate the trade-off between inflation and unemployment in the framework of a macroeconomic social welfare function. The ‘life satisfaction approach’ as a valuation technique has been applied to air pollution (Welsch, 2002, 2006), aircraft noise (Van Praag and Baarsma, 2005), climate (Rehdanz and Maddison, 2005), terrorism (Frey et al ., 2004) and fear of crime (Moore, 2006). In this literature, as in the present article, the terms happiness, life satisfaction, and subjective well-being are used interchangeably. To economists the terminology ‘experienced utility’ may sound more familiar, see Kahnemann et al . (1997).

³An alternative way of dealing with unobserved micro-heterogeneity is based on surveys in which a constant set of individuals is surveyed over time. This allows to use dummy variables for each individual as controls. The problem with this approach is that a fixed set of survey respondents will not remain representative across time.

⁴Note that the marginal rate of substitution is an ordinal concept, i.e. it is invariant with respect to the cardinalization of utility (happiness), as long as alternative cardinalizations are monotonically increasing transformations of each other.

⁵Since we follow the macro approach (using data on average happiness from representative surveys), personal and demographic variables (age, gender, marital status, etc.) play no role in this model.

⁶Age is an important predictor of individual happiness (see, e.g. Helliwell, 2003).

⁷The influence of corruption on capital productivity has been studied by Lambsdorff (2003). Consistent with the framework of this article, he established a negative impact.

⁸For simplicity, the notation e(v, w; c) omits the variables h and the controls.

⁹Note that the expenditure function could be extended to comprise more than just one production input, capital. The capital cost savings would then generalize to ‘factor cost savings’. In this article we deliberately treat labour input as fixed.

¹⁰Countries for which no data is available within 1998–2003 are disregarded.

¹¹The original data actually indicates absence of corruption, that is, 10 indicates a very clean and 0 a very corrupt country. In the present paper the scaling has been reversed such that 0 refers to the minimum and 10 to the maximum of corruption.

¹²See http://emlab.berkeley.edu/users/chad/HallJones400.asc

¹³Note that this variable plays a twin role: in the happiness equation it serves as a control that proxies the degree of rationality (modernity) of a society, whereas in the income equation it represents the human capital variable.

¹⁴Though unrealistic, we tested linear versions of the happiness equation and found the negative and significant linkage between happiness and corruption confirmed.

¹⁵For the model specified in Equations 8 and 9 the matrix of coefficients of the endogenous variables is triangular, implying that its determinant is 1. Thus, the Jacobian term in the loglikelihood function for the system (8), (9) vanishes, and the loglikelihood function has the same form as the loglikelihood function for a set of linear seemingly unrelated regressions (Davidson and McKinnon, 1993, pp. 644–645).

¹⁶I owe this observation to a referee.

¹⁷Consistent with the discussion at the end of subsection ‘Empirical approach’, the corresponding estimates in first differences yield much smaller coefficients of determination (Senhadji, 2000).

¹⁸The marginal significance level of corruption in the income equation is 1.29%, whereas it is 0.43 in the happiness equation.

¹⁹With respect to transition economies, May et al . (2002) report that corruption significantly enhances unofficial activity, whereas government effectiveness and the rule of law display no such influence.