In this research notes, based on a setting of online reputation system, we demonstrate various issues related to using regression discontinuity design (RD) to do causal inference, meaning to study whether for a given vendor, an increase in its online reputation score causes, instead of being associated with, an increase on its online sales. In the e-commerce platform, sellers are typically assigned into different reputation levels based on their original continuous reputation scores, resulting in discontinuity in reputation and allowing a regression discontinuity (RD) design setting. Based on data from one of the largest crowdsource website in China, we find that given everything else be the same, on average, sellers with one extra reputation level improvement enjoys a 32.5% increase in their sales. The RD required manipulation test indicates that both the density of the sales and various other shop related attributes show no discontinuity. Falsification test also indicates that the impact of reputation on sales holds only at true threshold. Further cross-sectional analysis show that the effect is stronger for seller with relative low reputation, owned by individual person, without the platform VIP membership or vendors not participating in the consumer guarantee program. This is consistent with the argument that new and unknown sellers depends more on the reputation system to improve their sales, offering further corroboration to our results.
