Date of Award
8-2026
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Economics
Committee Chair/Advisor
Babur De los Santos
Committee Member
Patrick L. Warren
Committee Member
Yichen Christy Zhou
Committee Member
Matthew S. Lewis
Abstract
Online review systems play an increasingly important role in reducing information frictions in digital markets. This dissertation studies three aspects of these systems, including sellers' incentives to manipulate reviews, the design of rating systems, and how horizontally differentiated negative reviews affect consumer demand. The first chapter investigates key drivers influencing online sellers' decisions to invest in fake reviews. I first construct a model in which sellers strategically invest in fake reviews to influence consumers, who form beliefs about product quality based on observed reviews. I then empirically test the model's predictions on the main determinants of fake review prevalence using Amazon product review data. To address the challenge of unobservable fake reviews, I use a machine learning model trained on verified fake Amazon reviews sourced from Facebook groups to estimate the likelihood of fake reviews for Amazon products. I find that products are more likely to have fake reviews when they are durable, are new, have a greater share of frequent consumers, or (in certain circumstances) face more substitutes. These findings help platforms and regulators better understand the economic environment to combat fake reviews.
The second chapter studies how the coarseness of rating systems affects the informativeness of online ratings about product quality. Our analysis combines two approaches. First, in a controlled experiment that varies vertical and horizontal differentiation, we find that the two-star scale better distinguishes product quality in a complex game, whereas the six-star scale provides clearer differentiation in a simple game. Second, using Yahoo! Movie and EachMovie data, we construct a benchmark for average perceived quality with film and viewer fixed effects and compare platforms’ raw rating rankings to this benchmark. The six-point scale aligns more closely with the benchmark than the thirteen-point scale. Together, the results suggest that rating scale design should be matched to consumers’ preference heterogeneity to ensure efficient information transmission and welfare.
The third chapter studies when negative reviews can increase product demand. While negative reviews are typically interpreted as signals of low quality, some instead reveal horizontal differences in consumer tastes. I exploit Amazon’s introduction of AI-generated review highlights as an information shock that makes review content easier to process. Combining Amazon review data with Keepa sales rank data, I compare Grocery and Gourmet Food products with high and low rating variance using a difference-in-differences design. High variance products improve their weekly sales rank by approximately 5,788 positions after the introduction of AI-generated review highlights. Event study estimates support the parallel trends assumption, and both rule-based and large-language-model classifications show that high variance products contain a larger share of horizontally differentiated one star reviews. The results suggest that negative reviews can raise demand when they reveal preference relevant attributes rather than vertical quality problems.
Recommended Citation
Yang, Huiyu, "Essays on Reputation and Feedback System in Digital Markets" (2026). All Dissertations. 4301.
https://open.clemson.edu/all_dissertations/4301