Assessing misspecification and aggregation for structured preferences

Applied research often tolerates misspecification in order to reach informative conclusions. We focus on how the degree of misspecification varies with the level of aggregation of data for quasilinear utility models. We present aggregation results formalizing that the model cannot get worse when aggregating. Using scanner data, we find that while all individuals are inconsistent with a quasilinear utility model, we cannot refute the hypothesis that a representative agent is a quasilinear utility maximizer. This provides evidence that deviations from a quasilinear model may average away.