Experiments on revealed preference often use budget sets that are randomly and independently drawn according to some criteria for each participant. However, this means that the budget sets faced by different individuals are not the same. This paper proposes a method to control for these differences. In particular, we control for the \power” of different budget sets by examining the consistency of an individual’s choices relative to some simulated baseline behavior conditional on budgets faced by the individual. We apply this methodology to two existing experimental datasets. Our results show that failure to account for this variation results in a bias when looking directly at measures of choice consistency and the sign of this bias depends on the measure being used. However, controlling for this variation does not change the correlation between measures of choice consistency and observable demographic characteristics like income and education.
(joint with John Rehbeck)