Uploaded October 2024 | Updated November 2026, 1 day ago
Tesary Lin – Assistant Professor, Boston University
Revealing Privacy Preferences: Experimental Insights on Consumer Data Sharing Decisions
Consumer privacy preference has long been an elusive topic for researchers and policymakers alike. A well-documented phenomenon is the “privacy paradox”, referring to the disconnect between stated attitudes and revealed preferences (i.e., actual behaviors) for privacy; additionally, the context-dependent nature of privacy preferences further complicates their measurement. In this talk, I present three experiment studies that tackle these challenges via various experiment design techniques. I demonstrate how to elicit revealed preference for privacy to ensure the managerial and policy relevance of the results, the role of theory in guiding experimental variation design, and how to implement user-side experiments in the field when the nature of the research question precludes industry-facilitated testing. I then discuss the insights these studies reveal about the nature of consumers’ privacy preferences.
David Choi – Associate Professor, CMU
Agnostic Characterization of Interference in Randomized Experiments
Mechanisms for interference–such as transmission of disease, sharing of information, peer influence, or economic competition–often play fundamental roles in our understanding of social behaviors, and their empirical characterization–such as their nature, prevalence, or strength–may be of scientific interest. In this work, we give a general approach for characterizing the prevalence of interference. This approach is applicable to experiments with binary-valued outcomes. It can be used to to explore questions such as Q1. How many units are affected by any treatment (including their own)? Q2. How many units are affected by the treatment of others? of distant others? Q3. How many units are affected by the treatment of others, provided that their own treatment satisfies some condition? For each of these questions, our approach gives conservative point estimates and one-sided confidence intervals, both of which lower bound the true value. Our point estimates are asymptotically equal to inverse propensity weighted (IPW) contrasts, such as comparisons of treated versus control, or comparisons of different levels of indirect exposure. Under reasonable experiment designs, our point estimates will converge to a lower bound on the true value, while the one-sided interval will have coverage probability converging to at least the desired level. These consistency and coverage properties will require no assumptions or restrictions on the nature of the interference, relying only on the known experiment design. As a result, our estimates will remain valid even if they use an observed social network that is only a crude proxy for the actual social mechanisms. We find empirically that the efficiency of our interval estimates (i.e., the interval widths) is competitive with, and often better than, that of the EATE, an assumption-lean treatment effect.
Tesary Lin – Assistant Professor, Boston University
Revealing Privacy Preferences: Experimental Insights on Consumer Data Sharing Decisions
Consumer privacy preference has long been an elusive topic for researchers and policymakers alike. A well-documented phenomenon is the “privacy paradox”, referring to the disconnect between stated attitudes and revealed preferences (i.e., actual behaviors) for privacy; additionally, the context-dependent nature of privacy preferences further complicates their measurement. In this talk, I present three experiment studies that tackle these challenges via various experiment design techniques. I demonstrate how to elicit revealed preference for privacy to ensure the managerial and policy relevance of the results, the role of theory in guiding experimental variation design, and how to implement user-side experiments in the field when the nature of the research question precludes industry-facilitated testing. I then discuss the insights these studies reveal about the nature of consumers’ privacy preferences.
David Choi – Associate Professor, CMU
Agnostic Characterization of Interference in Randomized Experiments
Mechanisms for interference–such as transmission of disease, sharing of information, peer influence, or economic competition–often play fundamental roles in our understanding of social behaviors, and their empirical characterization–such as their nature, prevalence, or strength–may be of scientific interest. In this work, we give a general approach for characterizing the prevalence of interference. This approach is applicable to experiments with binary-valued outcomes. It can be used to to explore questions such as Q1. How many units are affected by any treatment (including their own)? Q2. How many units are affected by the treatment of others? of distant others? Q3. How many units are affected by the treatment of others, provided that their own treatment satisfies some condition? For each of these questions, our approach gives conservative point estimates and one-sided confidence intervals, both of which lower bound the true value. Our point estimates are asymptotically equal to inverse propensity weighted (IPW) contrasts, such as comparisons of treated versus control, or comparisons of different levels of indirect exposure. Under reasonable experiment designs, our point estimates will converge to a lower bound on the true value, while the one-sided interval will have coverage probability converging to at least the desired level. These consistency and coverage properties will require no assumptions or restrictions on the nature of the interference, relying only on the known experiment design. As a result, our estimates will remain valid even if they use an observed social network that is only a crude proxy for the actual social mechanisms. We find empirically that the efficiency of our interval estimates (i.e., the interval widths) is competitive with, and often better than, that of the EATE, an assumption-lean treatment effect.



![BIG.AI@MIT 2024: Entrepreneurship and AI - Cooking Up the Future
Session Lead & Moderator:
Bill Aulet – Managing Director, Martin Trust Center for MIT Entrepreneurship
Panelists:
M.L. Carr – Partner, New Technology Ventures
Lily Lyman – General Partner, Underscore VC
Stephanie MacConnell – Director, MIT delta v NYC
Heidi Messer – co-Founder and Chairperson, Collective[I]
The question is not whether AI is coming but rather the question is how big, how fast and in what areas? While we might not be able to answer those questions, we do know that the future will be driven by entrepreneurs. Many of those entrepreneurs will be fueled by Venture Capital. In fact, today the dominant share of the all the venture capital funding being deployed is going to AI companies. In this session we will discuss what can we learn from the investment patterns evident from this tsunami. We will also look at this from an entrepreneur’s perspective and finally how AI is turbo charging the entrepreneurship process itself and the ramifications. All of this is cooking up a never before seen strange brew and in this session, we will take a peek in the kitchen to get some insights on the future. BIG.AI@MIT 2024: Entrepreneurship and AI - Cooking Up the Future](https://i.ytimg.com/vi/PxqsS2IaDN8/mqdefault.jpg)






