Time Cost Illusions: Purchase-Usage Gap in Online Education (with Jingcun Cao, Zengxiang Chen, Yufeng Huang)
Abstract: Using proprietary data containing complete consumer purchase and usage records from a large farmer education platform in China, we document a substantial purchase–usage gap: on average, consumers complete only 30% of the content they buy, despite an average course price of $74. We trace this gap to a time cost illusion: consumers underestimate the time cost of learning at purchase and thus choose courses that demand more effort than they can commit. More importantly, marketing messages that downplay effort amplify this bias. Over successive purchases, consumers partially correct the bias by shifting toward shorter, easier courses and becoming less susceptible to marketing messages. We build a structural model of purchase and usage in which marketing messages affect perceived time cost. Estimates imply that, at purchase, perceived cost is only one-third of the true cost and that marketing-induced illusion reduces consumer surplus by $48 per enrollment. Disabling effort-downplaying marketing messages lowers purchases by 12 percentage points but raises usage by 21 percent, as consumers self-select into lower-cost courses. Counterfactual simulations show that either (i) splitting long courses into shorter modules or (ii) disclosing historical completion rate reduces the purchase–usage gap by aligning ex-ante expectations with ex-post usage.Shaping the Influencers: The Role of Multi-Channel Networks (with Xiaojie Li)
Many influencers are not independent creators with full control over their content and business strategies. Instead, they are influenced by the influencer agencies (MCNs) they affiliate with. We leverage influencers' entry into and exit from these agencies on TikTok to examine how influencer agencies affect their content creation and monetization.3. Store Visits, Locations, and Customer Perceptions: Market Structure Analysis with Customer Trajectories in Shopping Malls (with Dai Yao)
Using 10 million customers' trajectory data from malls, we tailor a language (embedding) model to construct store co-visitation patterns, and decompose this measure into stated brand preference and physical location. We show how our metric improves the understanding of substitution patterns using store closure shocks and counterfactual location changes.4. Informal Ties and Sales Agent Turnover: Evidence from Randomized Dormitory Assignment (with Hongye Sun, Mitchell Lovett and Kun Peng)
Random dorm assignment groups sales agents with roommates sharing their hometown, dialect, or alma mater, giving exogenous variation in informal ties. Exploiting this setting, we show that informal ties are a first-order driver of turnover but have no discernible effect on performance. Selected work-in-progress
5. Influencer Collaborations (tentative title, with Guang Zeng and Liuyi Ye)
6. Information Transparency and Teamwork: Evidence from a Randomized Controlled Trial in Indian Schools (with Anuj Kumar and Xinzhi Rao)