Finding the Wise and the Wisdom in a Crowd: Estimating Underlying Qualities of Reviewers and Items

Consumers, businesses, and organizations rely on others' ratings of items when making choices. However, individual reviewers vary in their accuracy and some are biased – either systematically over- or under-rating items relative to others' tastes, or even deliberately distorting a rating. We describe how to process ratings by a group of reviewers over a set of items and evaluate the individual reviewers' accuracies and biases, in a way that yields unbiased and consistent estimates of the items' true qualities. We provide Monte Carlo simulations that showcase the added value of our technique even with small data sets, and we show that this improvement increases as the number of items increases. Revisiting the famous 1976 wine tasting that compared Californian and Bordeaux wines, accounting for the substantial variation in reviewers' biases and accuracies results in a ranking that differs from the original average rating. We also illustrate the power of this methodology with an application to more than forty-five thousand ratings of "en primeur" Bordeaux fine wines by expert critics. Those data show that our estimated wine qualities significantly predict prices when controlling for prominent experts' ratings and numerous fixed effects. We also find that the elasticity of a wine price in an expert's ratings increases with that expert's accuracy.