This paper analyzes which of various plausible determinants are significantly related to an index of "perceived corruption" compiled from business risk surveys for the mid-1990s. Using 2SLS to reduce problems of endogeneity and a variation of Leamer's "extreme bounds analysis" to test for robustness, it finds three factors robustly significant. Countries that were more economically developed and those which are former British colonies were rated "less corrupt". Those which have a federal structure were "more corrupt". Daniel Treisman Assistant Professor Department of Political Science University of California, Los Angeles 4289 Bunche Hall LA CA 90095-1472 Treisman@polisci.ucla.edu First Draft September 1997 Revised April 1998 ####

Theoretically, this article draws on political agency theory to formulate hypotheses. Empirically, it shows that political institutions have a role in explaining the prevalence of political corruption in American states. In the states, a set of democracies where the rule of law is relatively well established and the confounding effects of differing electoral systems and regimes are absent, institutional variables relating to the openness of the political system inhibit corruption. That is, other things equal, the extent to which aspiring politicians can enter and gain ®nancial backing, and to which voters can focus their votes on policies and thereby hold incumbent politicians accountable for policy outcomes and ®nd substitutes for them if dissatis®ed with those outcomes, reduce corruption as a general problem of agency. These institutional effects are estimated in the presence of controls for variables representing other approaches.

Charles Sampford and Carmel Connors (Griffith University) have asserted their rights under the