Research Proposal for: An Experiment on Online Ratings
Note: This presentation takes place before we run the main experiment. Here is what we plan to do:
We conduct an online experiment on Prolific that tests several features of online ratings. First, we provide additional robustness to existing results on polarization and welfare improvements. Second, following the spirit of (Acemoglu et al 2022), we quantify the selection effect of ratings, which results from only those who purchase a good being able to give a rating. We investigate how this effect impacts welfare and its relation to confidence. Third, we propose a new channel that may distort ratings: dependence. Most models assume that ratings are independent from previous ones. However, we expect to observe a tendency to overcorrect previous ratings in the direction of a customer's own evaluation. Finally, we address the issue of horizontally differentiated products. We first demonstrate that the welfare gained from rating systems is reduced compared to what we find for vertically differentiated goods. To solve this issue, we suggest grouping the ratings and showing them separately by the groups of relevance, similar to the policies adopted by some online marketplaces.