with Paul B. Ellickson and Mitchell J. Lovett

Online platforms are increasingly adopting multiple pricing display strategies, including partitioned pricing, shrouded pricing, and drip pricing, aimed at facilitating greater sales response. These non-transparent pricing techniques have drawn the attention of firms, policy makers, and consumer advocates. This paper empirically examines how consumers react to non-transparent prices that feature several distinct obfuscation strategies, and quantifies the effects of a policy aimed at increasing transparency. Using data on Airbnb property rentals in California, we empirically analyze the impact of different display strategies by leveraging two sources of variation in price transparency: (1) regime changes in the tax collection enforcement policy faced by the platform (which create identifying variation in the hidden components of price); and (2) a change in Airbnb's website display feature that creates variation in the degree of transparency across several pricing components. We find that consumers under-react to the hidden component of price: they act as if they ignore 70.5% of the hidden price under the old (less transparent) display regime. While consumers are more price-sensitive in the new (more transparent) regime, they continue to ignore 65% of the hidden price, indicating that the move to greater transparency was largely ineffective. To further isolate price responses on the partitioned  price components, we specify a structural model of limited consumer attention to quantify the impact of shrouded and drip pricing. We implement a semi-parametric pairwise maximum score estimation method that relaxes strong distributional assumptions and only requires comparisons between alternative pairs of choices. This estimator overcomes the challenges of a setting in which the number of alternatives is very large and key aspects of the choice set remain unobserved to researchers.

with Garrett A. Johnson and Scott K. Shriver, Marketing Science, 39(1)(2020):33-51. Winner, 2020 John D.C. Little Award

We study consumer privacy choice in the context of online display advertising, where advertisers track consumers' browsing to improve ad targeting. In 2010, the American ad industry self-regulated by implementing the AdChoices program: consumers could opt out of online behavioral advertising via a dedicated website, which can be reached by clicking the overlaid ``AdChoices'' icons on ads. We examine the real-world uptake of AdChoices using transaction data from an ad exchange. Though consumers express strong privacy concerns in surveys, we find that only 0.23% of American ad impressions arise from users who opted out of online behavioral advertising. We also find that opt-out user ads fetch 52.0% less revenue on the exchange than do comparable ads for users who allow behavioral targeting. These findings are broadly consistent with evidence from the European Union and Canada, where industry subsequently implemented the AdChoices program. We calculate that the inability to behaviorally target opt-out users results in a loss of about $8.58 in ad spending per American opt-out consumer, which is borne by publishers and the exchange. We find that opt-out users tend to be more technologically sophisticated, though opt-out rates are also higher in older and wealthier American cities. These results inform the privacy policy discussion by illuminating the real-world consequences of an opt-out privacy mechanism. 

with Paul B. Ellickson and Mitchell J. Lovett

Grocery shopping, a traditionally offline category, has experienced rapid growth over the recent years with the thriving of many online grocery retailers entering the market. We examine whether and how consumers' shopping habits are affected by purchasing groceries online for the first time. Using NielsenIQ Consumer Panel data from the Kilts Center for Marketing, we measure the spillover and substitution effects for the consumers between online and offline channels and across retailers to gain a complete view of consumers' short-term and long-term reactions. To address the concerns about self-selecting into online grocery shopping treatment, we implement a recently advanced method of causal inference with machine learning to generate synthetic counterfactual control households comparable to the treated counterparts using an interactive fixed effect model. This model accounts for latent time-varying confounding factors correlated with self-selecting to be treated. We utilize two different estimators to estimate the model and compare their performances: generalized synthetic control and matrix completion. Results from both estimators show that the average treatment effect is significant only in the short-run (within six months): consumers increase their shopping trips and spend $7.95 more on food products, which is equivalent to a 10% of their weekly food expenditure. However, there is no significant treatment effect in the long run. These findings suggest that, having trialed online grocery shopping, consumers tend to re-allocate purchases across retailers rather than increasing their overall engagement with online shopping.