Data-Driven Price Discrimination and Data Manipulation: A Laboratory Experiment

How to cite

Wang, Jianyun. 2026. "Data-Driven Price Discrimination and Data Manipulation: A Laboratory Experiment." AEA RCT Registry. AEARCTR-0018875. https://doi.org/10.1257/rct.18875-2.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Jianyun Wang*

* Primary investigator

Registry ID
AEARCTR-0018875
DOI
https://doi.org/10.1257/rct.18875-2.0
First registered
Last updated
Status
In development
Trial dates
2026-06-09 – 2026-11-20
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/18875

Abstract

This laboratory experiment investigates the strategic interactions between a digital platform and consumers, testing the theoretical boundaries of data-driven price discrimination and consumer data manipulation.
In a controlled laboratory setting, we implement a game between consumers and the platform. Consumers (played by subjects) are assigned types and choose their platform usage levels (denoted by action T and action M), which generates behavioral data. Simultaneously, the platform decides on its unobservable data-processing investment(i), which determines the probability of successfully decoding usage data for personalized pricing. If data processing succeeds, the platform offers a personalized price to the consumer; otherwise, an anonymous uniform price is charged based on the prior distribution of types.
Our experiment specifically tests three key theoretical predictions:
Strategic Obfuscation (The Mimicry Effect): Whether and how many high-type consumers strategically reduce their usage to pool with low types when the platform's investment (i) is uniform to all consumers, thereby compressing the informational value of the data.
Heterogeneous Risk Preferences: How consumers' risk-aversion levels (r) scale their incentives for strategic data manipulation and alter the welfare-optimal allocation of data rights.
The Welfare effect of subsidy: Consumers who choose action M, announce a minimal subsidy that make them willing to choose action T. How such subsidies affect the platform's investment and the social welfare.

Experimental design

### Project Overview and Experimental Design

This study employs a two-phase lab experiment to investigate consumers' strategic data manipulation (misrepresentation) and their valuation of data privacy under personalized pricing.

#### Phase 1: Risk Elicitation and Cognitive Screening (Prior Calibration)
We will recruit approximately 120 subjects to participate in 4 independent sessions (30 subjects per session). This phase consists of two tasks:
1. **Risk Preference Elicitation:** Subjects complete the Bomb Risk Elicitation Task (BRET) to measure their individual risk aversion parameter (\alpha).
2. **Cognitive Ability Test:** Subjects complete a brief cognitive reflection and numeracy test to screen for comprehension and ensure sufficient mathematical ability.
The individual data collected in Phase 1 will be used to calibrate the theoretical parameters and empirical distributions for Phase 2.

#### Phase 2: The Main Interactive Experiment
Subject to attrition, we expect 60 to 90 subjects to return from the Phase 1 pool, who will be randomly grouped into 3-player triads. Each triad consists of one Consumer A (High Type), one Consumer B (Low Type), and one Platform. While subjects are assigned to these roles, the Platform's investment decisions are programmatically determined by the computer at the theoretical equilibrium values.

Phase 2 consists of two consecutive parts:

##### Part 1: Repeated Dynamic Game (20 Rounds)
For 20 rounds, subjects interact in their fixed triads. In each round:
- **Consumer's Decisions:** Consumers can choose either "Action T" (truthfully revealing their assigned private type) or "Action M" (manipulating their data to misrepresent themselves as the other type).
- **Belief Elicitation:** Consumers are also asked to predict the platform's investment level.
- **Platform's Strategy:** The platform's investment is simulated by the computer program and fixed at the theoretical equilibrium value.
- **Payoffs:** Individual round payoffs are jointly determined by the consumer's decision (T or M), the computer's investment level, and a random stochastic factor.
- **Incentives:** At the end of Part 1, one of the 20 rounds will be randomly selected for actual payment.

##### Part 2: Elicitation of WTA via BDM Mechanism (1 Round)
Roles from Part 1 remain strictly unchanged. Part 2 lasts for exactly 1 round:
- **Low Type (Consumer B):** Performs the same truth-telling/manipulation decision as in Part 1.
- **High Type (Consumer A):** Instead of a binary choice, Consumer A utilizes the Becker-DeGroot-Marschak (BDM) mechanism to bid a minimum price (subsidy) they are willing to accept to choose "Action T" (truthful revelation). Theoretically, Consumer A who voluntarily chose T in Part 1 should bid 0, while those who chose M in Part 1 should bid a positive WTA.
- **Platform's Strategy:** The investment level remains computer-generated.
- **Incentives:** This single round is 100% paid and added to the subject's final cumulative earnings.

Design details

Randomization method
Randomization done in office by a computer.
Randomization unit
Individual level: Within each experimental session, subjects are randomly assigned to roles (Consumer A, Consumer B, or Platform) by the computer software.
Clusters
20 to 30 clusters. This study consists of two phases. Phase 1 (Risk Elicitation) will recruit 120 subjects who make individual decisions. Phase 2 (the main interactive experiment) will invite the same pool of 120 subjects back. Accounting for a predicted 25% to 50% attrition rate between phases, we expect 60 to 90 subjects to return for Phase 2. Since subjects in Phase 2 are grouped into 3-player triads (Consumer A, Consumer B, Platform), this will yield between 20 and 30 independent clusters.
Observations
20 to 30 independent group-level observations for the primary game, plus 120 individual-level observations for the prior risk distribution calibration.
Arms
This is a two-phase, within-subject design with no separate treatment arms across subjects: - Phase 1 (Prior Calibration): Target sample of 120 individual subjects to elicit risk aversion parameters (r). This distribution will be used to calibrate the theoretical parameters for the subsequent game. - Phase 2 (Main Experiment): Target sample of 20 to 30 active clusters (60 to 90 returning subjects). All 20 to 30 active clusters in Phase 2 will experience both the baseline condition (no subsidy elicitation) and the treatment condition (subsidy elicitation via BDM) dynamically within their respective play. There is no between-subject division; all returning subjects participate in the same integrated within-subject environment.
Primary outcomes
The choices of conusmers. The subsidies that consumers announce through BDM mechanism. The aggregate social welfare.

Cited by

No citation data available for this registration yet.

Keywords