A non-parametric test of stochastic preferences

In this paper we present algorithms to test a theory of stochastic preferences on binary choice data. For the general case, where preferences can be any strict linear order, a column generation algorithm is given. This algorithm can be easily adjusted to test for specific classes of preferences or to account for different kinds of data. We perform preliminary experiments showing the viability of this method. Furthermore, we look at the special case of single-peaked preferences and show that for this class, a test of stochastic preferences can be done in polynomial time by checking the data for a simple condition.status: publishe