Social Sciences Brown Bag Seminar

Revealed preference tests are elegant nonparametric tools that ask whether individual or aggregate data conforms to economic models of optimizing behavior. In designing a test using revealed preference, however, one faces a vexing tension between goodness-of-fit and power. If the test finds violations, then one must ask if the test was too demanding-is there an acceptable tolerance for goodness-of-fit? On the other hand, if no violations are found, one must demonstrate that the test was demanding enough-is the test sufficiently powerful? This paper provides a counter-weight to the many papers on goodness-of-fit by discussing indices of the power of revealed preference tests. By focusing on the underlying probability model induced by sampling consumers from the population, we attempt to unify the two approaches. We present several new indices and discuss their relative merits, illustrated by applications using experimental data.

http://hss.caltech.edu/~bgillen/papers/GARPPower.pdf