D RA FT : D o no t c ite . Facts , Perceptions or Survey Questions : How should we measure corruption ?

Corruption is an important economic variable but one that is extremely difficult to measure. As a result, researchers have resorted to using perception indexes as a proxy for true corruption. Recent studies suggest that perception measures are inaccurate, and have encouraged the use of methodologies that can measure fact-based corruption – corruption that actually takes place. In this paper, we argue that fact-based corruption measures are often worse indicators of true corruption levels than are perceptions. To show this, we conduct a laboratory experiment where we exogenously vary the true level of corruption (as measured by the frequency and size of bribes taken from investors), and we compare three different measures of corruption: one based on facts, another based on perceptions, and one based on survey questions asking about facts. We find that all three measures correlate well with true corruption, with perceptions doing the best. We also find that all measures suffers from biases that can make move them opposite that of true corruption. ∗We would like to thank Kristine Brown, Cary Deck, Ori Heffetz, and the graduate students in Experimental Economics. Xirong Chen provided excellent research assistance. We thank the Sutton Fund and the University of Arkansas for financial support. Contact Info: Department of Economics, Sam M. Walton College of Business, Business Building 402, Fayetteville, AR 72701-1201, USA. Email: [email protected], [email protected]