Four essays on modelling and estimating consumer heterogeneity in probabilistic choice and household demand systems

摘要

This thesis consists of four essays on modelling and estimating unobserved consumer heterogeneity in microeconometric models of probabilistic choices and households demands. Incorporating such heterogeneity into these models potentially allows the researcher to infer more details on the underlying decision processes from the observed data. Each essay explores the extent to which estimated heterogeneity can help developing answers to a research question and improve the reliability of empirical findings. The first essay focuses on interpersonal heterogeneity in attribute non-attendance. Several studies estimate its incidence using a latent class logit model with pre-specified support points wherein the share of each point corresponds to the probability of ignoring particular attributes. This essay shows that because those points can be used to approximate unmodelled taste variation empirically, the interpretation of estimated shares is necessarily ambiguous. The second essay advances a new perspective on the problem of unstable coefficients in the rank-ordered logit model. This problem has been traditionally explained by intrapersonal heterogeneity, specifically that individuals rank less preferred items more erratically. This essay shows that the problem may originate instead from the model's sensitivity to stochastic misspecification in the postulated random utility function. Even a form of misspecification which is mostly inconsequential for the multinomial logit model can affect the rank-ordered logit model discernibly. The third essay compares stated preferences over nursing job attributes elicited by traditional and new types of discrete choice experiments, involving choices over jobs and attribute-levels respectively. Respondents are found to place greater value on pecuniary over non-pecuniary gains when completing the traditional choice task. The estimated pattern of taste heterogeneity suggests that the discrepancy cannot be explained by the typical conjecture that the more complex traditional task induces heuristic choices. The fourth essay explores the association between disability and unobserved intra-household expenditure allocation processes in a collective framework. This association may yield a refined measure of extra living costs resulting from disability. An analysis using Canadian expenditure data shows that the structural identification strategy in the state-of-the-art collective model does not generate robust quantification of this association.