Selected Research

The Consumption of Serial Media Products and Optimal Release Strategy
(Job Market Paper)

with Nitin Mehta and Mengze Shi

We study the impact of a digital platform’s product release strategy of serialized media content on its consumption and the ensuing profits made by the platform. Specifically, we study in the context of a digital book platform, whether a platform should release all the chapters of book simultaneously to its consumers or should it release it sequentially over time, or should it follow a hybrid strategy where it releases a fraction of the chapters of a book simultaneously and the rest sequentially. We show that the release speed of chapters impacts consumption through two opposite forces: binge consumption and product exploration. A slower release speed of chapters, while limiting binge consumption, leads to an increased exploration of other books through a combination of two mechanisms - consumers visit the platform more often and the explore other books due to the constrained availability of the focal book. We estimate our structural model on consumers’ purchases of chapters on an online book platform in China, which currently uses a sequential release strategy. Our counterfactual analysis shows that the platform will be worse off with a simultaneous release of chapters but it can improve its sales revenue by implementing an optimized hybrid release strategy.

Incentivizing Mass Creativity: An Empirical Study of Online Publishing Market
with Xiaolin Li and Mengze Shi [Draft]

  • Recipient of 2019 China Research Grant

This paper examines the effects of quantity-based incentive plans on the quality of creative production. We study a publishing platform which switched from a uniform commission rate to a quantity-based plan offering higher commission rates if a writer's production meets higher quantity brackets. Theoretical analysis indicates that a quantity-based commission plan can enhance the quantity-quality complementarity: the creators who reach a higher bracket of quantity should also produce a higher quality. This theoretical result is confirmed consistently in multiple empirical tests. First, for a given book, the chapters published in the months when writers reached higher brackets of quantity had higher quality measured by chapter-to-chapter customer retention rates. Such a positive correlation is not significant in books published when the platform offered a uniform commission plan. Second, when writers produced higher quantity in the early days of a month but failed later, there was a greater quality drop under the quantity-based commission plan than under the uniform commission plan. Finally, in response to positive demand shocks, writers reached higher quantity brackets and increased quality. Overall, our results underscore the importance of proper incentive design in managing mass creativity.

How Do People Update Beliefs? Evidence from the Laboratory
with Andrew T. Ching, Tanjim Hossain and Shervin S. Tehrani

The Bayes' rule serves as a cornerstone for learning models when researchers observe consumer choice, but not underlying beliefs. Observational data usually provide information on consumer choices and the environment but lacks information on consumer beliefs. As a result, the vast majority of the existing research assume Bayesian updating process. In this paper, we design a laboratory experiment to directly elicit subjects' posterior beliefs as she receives more information about a random variable in a dynamic setting. We elicit both mean and the uncertainty associated with it and investigate subjects' belief updating rules without assuming strict adherence to the Bayes' rule. We find that subjects are heterogeneous in their belief updating rules. Relative to Bayesian updating, some subjects are too conservative and others are overly reactive to new information signals. Moreover, as subjects receive more signals, their updating rules slowly change over time. Nonetheless, subjects of all types become more reactive to new signals and become more certain about their beliefs as they receive more signals. In addition, subjects assign greater weight to surprising signals and their beliefs become more diffused.

A Neuro-Autopilot Theory of Habit: Evidence from Canned Tuna
with Colin Camerer, Peter Landry, Matthew Osborne and Ryan Webb [Draft]

  • Recipient of 2019 SSHRC Insight Development Grant

The existing marketing and economic literature on consumer choice typically models habits by allowing the preference to change based on past consumption. While this approach is able to capture choice persistence, it ignores the role of attention and effort reduction in habitual decision making. In this paper, we propose that habit represents one of two distinct decision-making modes. The habit mode automatically repeats past choices with minimal deliberation. The other mode involves conscious deliberation over all available options. The transition between these decision modes is governed by the reliability of a reinforcement learning algorithm. We provide empirical evidence for this concept of habit by analyzing consumer choices in the canned tuna category around 2008, when the major brands shrunk the size of their packaging in the US market from 6oz to 5oz per can. Our structural model results show that our neural-autopilot model of habit largely improves the model fit relative to the standard consumer choice model with state-dependent utility. We find strong evidence for choice persistence arising from a habitual autopilot decision mode – 91% of canned tuna choices are made in habit mode. These choices are not “optimal” in the sense of using all available information, but they do not require cognitive effort. The effect of past choices on utility, i.e., the state-dependency parameter estimate, is halved after the model allows for autopilot. This indicates that state dependence may play a smaller role in choice persistence than estimated in previous literature.