This thesis points to a possible weakness of the empirical literature on corruption and government spending. That corruption affects the composition of government spending, and in particular that it affects education and health spending adversely, seems to be empirically well established. However, there exist additional literature closely related to corruption and government spending, treating(i) a relationship between corruption and decentralization, and (ii) a relationship between decentralization and government spending. These relationships are not accounted for in the literature on corruption and government spending. If corruption and decentralization are correlated, and in addition decentralization affects government spending, then omitting decentralization might causebiased results.
In order to test for possible omitted variable bias, a simple cross-country analysis is performed. Three versions of the classical linear regression model are specified,and estimated in Stata 10 using ordinary least squares estimation. In the first version, the relationship between corruption and government spending, excluding decentralization, is investigated. In the next version, the model is extended to include decentralization as explanatory variable, while in the third version, an interaction term between corruption and decentralization is included as well. These models are then estimated using different measures of government spending on education and health as dependent variables.
The results support the findings that corruption adversely affects government spending on health and education. These findings seem to be very robust, also when decentralization is included as an explanatory variable. Nevertheless, there are indications that there might be an interaction effect between corruption and decentralization. Specifically, government spending on health seems to be affected more negatively by corruption in decentralized countries. There is a need to investigate this further. In addition, an explanation of this interaction effect remains to establish. Meanwhile, an intuitive guess is that corruption can take different forms at different government levels, and that these different forms of corruption may have different effects on health spending.