This website uses cookies.

By clicking the "Accept" button or continuing to browse our site, you agree to first-party and session-only cookies being stored on your device to enhance site navigation and analyze site performance and traffic. For more information on our use of cookies, please see our Privacy Policy.

  1. Home
  2. ASSA Annual Meeting
  3. 2027
  4. Preliminary Program
  5. Platforms and Digital Economy
« Back to Results

Platforms and Digital Economy

Paper Session

Sunday, Jan. 3, 2027 8:00 AM - 10:00 AM (EST)

Walter E. Washington Convention Center
Hosted By: Chinese Economists Society
  • Chair: Qihong Liu, University of Oklahoma

Algorithms, Biases, and Belief Polarization

Varad Deolankar
,
National University of Singapore
Jessica Fong
,
University of Maryland
S Sriram
,
University of Michigan

Abstract

Algorithmic curation by social media platforms is often blamed for rising polarization. Yet intrinsic biases in how humans process information, such as signal distortion (perceiving information as more aligned with one's belief) and weight distortion (discounting belief-incongruent information), can also contribute. This paper investigates the relative contributions of these two pathways and examines how the engagement metric the algorithm optimizes, such as dwell time or likes, shapes outcomes. The four-step approach combines (1) an online experiment tracking belief updating and engagement, (2) model estimation, (3) deep reinforcement learning agents trained on the estimated model to emulate personalized recommenders, and (4) counterfactual simulations of posterior belief distributions. The findings show that information-processing biases and algorithmic curation each amplify polarization to a similar degree. Among biases, weight distortion is the primary channel. Because individuals dislike opposing content yet spend more time reading it than confirmatory content, like-maximizing and dwell-time-maximizing algorithms curate divergent feeds. Both increase polarization relative to random assignment, but dwell-time maximization leads to less polarized beliefs than like maximization.

Learning from the Customer Journey in Online Shopping

Liangzong Ma
,
Harvard Business School
Rajiv Lal
,
Harvard Business School
Shunyuan Zhang
,
Harvard Business School

Abstract

Purchases and returns in e-commerce are not one-time decisions — they are shaped by the entire customer journey leading up to them. Most existing approaches focus on static information, such as past purchases or overall activity levels, and overlook how customers actually move through the site step by step. This paper uses detailed clickstream data from a large apparel retailer to study how the sequence of customer actions such as browsing or searching can help predict both purchases and returns. Comparing traditional models with models that take the full journey into account, the results show that using the full sequence of customer behavior improves prediction for both purchase and return decisions. Purchase intent and return risk develop differently over time, and journey information is especially useful for customers with little historical data. Overall, the results suggest that understanding how decisions unfold over time can help firms better identify when customers are uncertain, improve decision quality, and reduce costly product returns.

Compensation Transparency and Worker Moral Hazard Behavior on Digital Platforms

Mengwei Qu
,
University of Connecticut
Chen Liang
,
University of Connecticut
Chunxiao Li
,
University of Science and Technology of China
Bin Gu
,
Boston University

Abstract

Digital platforms increasingly rely on transparency policies to govern gig workers, yet the behavioral consequences of such policies remain unclear. In ride-hailing, drivers often observe their own earnings but not how passenger payments are allocated between drivers and the platform, creating uncertainty about whether the platform's commission is fair. Payment-allocation transparency may reduce moral hazard by resolving this information asymmetry, but it may also reveal unfavorable allocations and make fairness comparisons more salient. Leveraging trip-level data from the staggered rollout of a payment-allocation transparency policy on a leading ride-hailing platform, we estimate a difference-in-differences model. We find that transparency reduces drivers' moral hazard, measured by detours and fare padding. However, this governance effect depends on the fairness information the policy reveals. When drivers receive a lower payment share relative to the platform or to peer drivers, transparency's moral-hazard-reducing effect weakens. We further find that the effect is weaker among team-managed drivers, suggesting that informal team governance can partially substitute for formal transparency policies in disciplining driver behavior. These findings show that transparency is not merely an information-disclosure tool: it shapes how workers evaluate payment allocations, compare outcomes, and respond to platform governance.

A Study in Entry: Platform Location Choice in Embedding Space

Ben Leyden
,
Cornell University

Abstract

Where do platform owners enter the product space they host? When Apple, Google, or Amazon launch products on the marketplaces they run, those location decisions shape competition with complementors and the direction of third-party innovation. Studying this empirically has been hindered by measurement: a platform's products overlap continuously, and any market boundary drawn through that continuum is a researcher's choice the analysis then inherits. Treating the product space as continuous, this paper uses sentence-transformer embeddings of product descriptions to locate each product, and defines neighborhood outcomes — density, quality, prominence, demand growth, and others — via Nadaraya-Watson kernel interpolation. Using dartboard permutation tests, a conditional logit model of entry choice, and a temporal panel of entry timing, the paper studies where Apple and Google enter across the iOS App Store and Google Play. Whether the two platform owners enter alike bears on whether self-preferencing rules should govern one firm or gatekeepers as a class.

Discussant(s)
Guangying Chen
,
University of Southern California
Lan Luo
,
Yale University
Anya Shchetkina
,
Massachusetts Institute of Technology
Ginger Jin
,
Boston University
JEL Classifications
  • L8 - Industry Studies: Services
  • D8 - Information, Knowledge, and Uncertainty