Three Essays on Reference-Dependent Choice

Kahneman and Tversky (1979) illustrated that decision-makers tend to judge stimuli relative to some reference point, whereby consumers react stronger to framed losses than gains. While the two concepts of reference dependence and loss aversion have been proved to hold in many research contexts, we still know little about how the reference-dependence paradigm works in many product choice situations (Mazumdar, Raj, and Sinha 2005). Bridging behavioural and marketing sciences, this dissertation tackles several relevant questions of reference-dependent choice in three essays. The first thesis chapter provides a quantitative review of loss aversion in product choice. To achieve this, we conduct a meta-analysis of studies that implemented reference dependence in random utility brand choice models. This meta-analysis sheds light on the robustness and presence of the loss-aversion phenomenon across choice contexts. Specifically, we use multilevel modelling techniques to examine potential moderators of loss-gain asymmetries as well as the variability of loss-aversion effects within and between studies. We find that loss aversion is manifest in product choice, but that it exhibits substantial variation across research contexts.Because nearly all studies on reference-dependent preferences centre on nondurable products, Chapter Two of the dissertation examines in an experimental setting how consumers choose durable goods from presented assortments and what role reference dependence plays in this situation. To empirically test which reference-point theories best describe consumer behaviour, we develop and estimate a latent-class mixed logit model that accommodates both taste and structural heterogeneity. In this background, we demonstrate the theoretical and practical importance of considering the diverse product-evaluation processes consumers follow to pick a product from a focal choice set. In the third and final chapter of the thesis, we further investigate the adoption decision of durable goods using revealed preferences. Since panel data are typically sparse for durable products, we estimate logit models of demand using aggregate sales data (Berry 1994; Berry, Levinsohn, and Pakes 1995). In this context, we illustrate the explanatory power of model formulations that consider reference dependence as well as extremeness-aversion context effects (Simonson and Tversky 1992)