The mechanisms through which permanent income affects bribery are remarkably similar in the disparate settings of Peru and Uganda. The rich bribe more frequently than the poor, principally because they use more officials than the poor and, in Peru, use a more corrupt mix of officials. Amongst users of officials, the rich are slightly more likely to bribe than the poor, although the gap is smaller when unwitting bribes are taken into account. However, although rich bribers pay larger bribes, consistent with price–discrimination, the bribes are a smaller share of their income than is the case for poor bribers.
. . .
The main benefit of bribery to a client is avoidance of the poor service delivered to clients who refuse to bribe. The results suggest that service improvements in response to a bribe merely offset service reductions associated with angling for a bribe, and that clients refusing to bribe are punished.
Our results suggest that the main costs of bribery lie in efficiency losses, as any large distributional effects must come indirectly through the performance of the economy. Nevertheless, the Ugandan data indicate that a good starting point for reducing bribery for the poor would be to reduce unwitting bribery by improving literacy and clearly publi- cizing official costs for services. More generally, the results highlight the power of public officials in their relationship with clients and the importance of weakening this power, for example, by providing clients with a choice of official, and rotating officials through jobs.The paper, "Is Bribery Really Regressive? Bribery’s Costs, Benefits, and Mechanisms," is by Jennifer Hunt and Sonia Laszlo. It is published in the latest number of World Development. A 2007 draft is here.
A very interesting new paper on the economics of bribery is here - it compares the efficiency of two punishment systems in Germany and China. It has useful policy implications.