Economics at a crossroads • Traditionally based on choice behavior • Choices to be respected – consumer sovereignty • Consistent choice explained by a utility representation • Comparable utility representations lead to consistent normative evaluations of policies The rational choice method has colonized political science, parts of sociology and is standard in 95% of economic analysisAt the same time…. • Accumulating evidence that choice is not consistent Inconsistency as: • Random error • Biased error • Weakness of will • Multiple unreconciled objectives/drives of biological subsystems  Fitness for bygone environments? • Presentation and context effects Beneficial policy evaluation based on consumers’ sovereignty: Uncover and estimate a latent preference despite inconsistent observed choicesDifferent forms of inconsistency in simple problems and complex problems Simple problems: • We (outsiders) know what is best, and reflection by subject will generally agree. • Benign paternalism can be useful in generating compliance. Complex problems: Too hard for outsiders. Actual decision making • A composition of many small complex problems • Appear, in the aggregate, as a larger simpler problem. Needs to be solved “on the ground”Repeated or increasingly detailed questions about a choice problem may not reveal the same preference relation as “natural conditions” • Reasons • Finding one’s own way • Time to explore counterfactuals Non-stochastic true preference: • Known to subject • Unknown to subject Introspection costly or free Intervention: • Passive May have less influence on natural choice • Active May help subject understand a complex problem or articulate reasonsStochastic true preference • Time scale may make a stable preference relation appear stochastic Preference for “variety” Or, • The preference may really be stochasticStochastic preference: • How to distinguish stable random preference from error-ridden preference Conditions for random choice to be consistent with an underlying stochastic preference relation: Falmagne (1978) J. Math. Psych. Barbara-Pattanaik (1986) Ecm.Falmagne-Barbara-Pattanaik (FBP) test is based on “full observability” x alternatives X all possible alternatives A available set Data is c(x,A) = prob of selecting x from A All (x,A) are in the domain. Define qc recursively by: qc(x,φ) = c(x,X) If not xεA: qc (x,A) = c(x,X\A) – Σ B in A qc (x,B) If xεA: qc (x,A) = 0 FBP Necessary and Sufficient conditions for a rationalization by random preference: qc (x,A) non-negative for all x,AThree questions: 1. Need to have a method for testing the Falmagne conditions without enumerating (x,X). CS methods for restricting stochastic preference or using the structure of X are needed to make this computationally feasible. 2. Need to have a non-parametric estimate of stochastic preference based on limited data. In what cases can this be used as a stochastic latent preference? 3. What to do when FBP test fails on limited data set, even under the appropriate CS restrictions of