Estimating Choice Models with Unobserved Expectations over Attributes

Agents often choose based on expectations about uncertain attributes, yet researchers rarely observe the information behind those expectations. We develop two methods to estimate discrete choice models with heterogeneous, unobserved information sets. Our point-identification approach uses repeated realizations of the uncertain attribute to estimate its statistical process with observed predictors and recoverable latent factors, then represents conditional expectations as a finite mixture over these components. A complementary approach set-identifies preferences using partial information about expectations. Both methods apply to individual- and market-level data. Applications to firms’ export decisions and consumers’ automobile purchases show that our method corrects for substantial bias in preference estimates that arises when unobserved information heterogeneity is ignored.