Bayesian analysis of consumer willingness-to-pay

This dissertation examines three problems in the use of discrete choice models to estimate consumer willingness-to-pay (WTP). First, we examine two parameterizations of choice models to estimate WTP. For maximum-likelihood estimation of homogenous and latent-class versions of the models, WTP estimates are invariant to parameterization. We show that Bayesian estimates are not strictly invariant. In sparse data conditions, the WTP estimates can be sensitive to parameterization. In this setting, indirect identification can lead to implied WTP distributions with poorly behaved or even undefined moments. Second, we study the WTP for intangible product attributes. We present a Bayesian factor-choice model to estimate the WTP for brand image associations. The factor model accounts for the binary nature of the manifest variables while the choice model is parameterized to directly estimate the WTP for the brand image factors. We estimate the model parameters simultaneously, avoiding the attenuation bias from sequential estimation. We find that the brand image data explains much of the unobserved heterogeneity in the distribution of WTP for the brands in our application. Third, we study the correlation in preferences for private label brands across unrelated categories of consumer packaged goods. We derive a multi-category choice model from consumer surplus maximization, which directly identifies the WTP for the private label brand. A variance components approach is used to handle the dimensionality problems inherent to multi-category models. This allows us to parsimoniously compute the WTP for the private label brand across categories. We find the correlation in private label intercepts computed from a conventional linear utility model overstates the correlation in private label valuation due to the correlation between the intercepts and price sensitivity.