Random Utility Models, Wine, and Experts

Abstract

By conducting a field experiment, we investigate whether consumers value expert opinion labels on wine as a form of reducing asymmetric information about product quality. We use two types of data. First, we use a macro-level monthly-product-store dataset, collected before and after our field experiment, which involves treating a random subset of wine products by displaying expert scores in one store, and comparing sales with wine sales in similar, non-treated stores. Secondly, we use a micro-level panel dataset from the treated store that provides information on products purchased and household characteristics. We combine these data with additional information on products, such as, varietal, region of production, price point relative to other wines, and expert scores. Using the macro-level data for treated and control stores, we estimate a structural random coefficient demand model for wine. With the household micro dataset, we estimate a random coefficient mixed logit demand specification, allowing for consumer heterogeneity in demand for wine. In order to capture the demand for wine, the products are defined as bundles of attributes, including the expert score that is experimentally introduced into the market. We find robust results in terms of consumer valuations for expert scores, using both datasets. In particular, we obtain an implied average willingness to pay (WTP) between 2 (using the macro level dataset for treated and control stores) and 3.2 dollars (using the micro level dataset for the treated store) for an average score of 83. Although not all of the consumer demographics help explain the heterogeneity in the value of expert scores, the wine ratings matter significantly less to consumers that have higher incomes, or are more likely to own a home. Finally, using counterfactual simulations we estimate the changes in consumer surplus resulting from available quality information in the form of expert opinion scores. Using the micro level dataset we find that removing scores leads to significant welfare losses especially for lower income consumers and for men. Overall, using the macro level dataset for treated and control stores, we estimate there to be a significant welfare loss eliminating scores of roughly one percent of total wine revenue in the treated store.

Replaces

Céline Bonnet, James Hilger, and Sofia B. Villas-Boas, WTP 4 WEO, February 28, 2017, revised June 2020.

Reference

Sofia B. Villas-Boas, Céline Bonnet, and James Hilger, “Random Utility Models, Wine, and Experts”, American Journal of Agricultural Economics, vol. 103, n. 2, March 2021, pp. 663–681.