Michael Keane, Jonathan Ketcham, Nicolai Kuminoff and Timothy Neal
Abstract: We propose new methods to model choice behavior and conduct welfare analysis in complex environments where it is untenable to assume that choices fully reveal preferences. In particular, we investigate how Medicare beneficiaries choose prescription drug plans (PDPs) under the Medicare Part D program. Our approach is novel in that we estimate a multinomial logit model for PDP choice that allows for heterogeneity in both preferences and the behavioral choice process. We find the data can be well characterized by a mixture of three behavioral types: The “rational” type constructs expected out-of-pocket costs E(OOP) rationally, and, ceteris paribus, seeks to minimize premiums plus E(OOP) as theory suggests. The second type constructs expected out-of-pocket (OOP) costs rationally, but puts too much weight on premiums relative to E(OOP) in choosing plans. A third type, who we label “confused,” places weight on irrelevant financial aspects of drug plans, implying they fail to construct E(OOP) rationally. A consumer is more likely to be the “confused” type if they suffer from Alzheimer’s disease and/or depression. We use the model to quantify the monetary and welfare losses that arise from suboptimal decision making for the population, for the behavioral types, and for people with cognitive limitations. We also evaluate policies to simplify the choice set to reduce these losses.
Keywords: Random utility model, Mixture of experts, Mixed Logit, Market mapping, Hedonic Utility, Decision utility, Medicare, Health insurance, Behavioral economics