Discrete Choice with Generalized Social Interactions
summary
The gist
Text A presents an abstract/summary of a theoretical model concerning social interactions in discrete choice models…
In short
This work develops a discrete choice model to study how social interactions affect individual decisions, allowing agents to conform or differentiate from network members. The authors solve identification problems caused by unobserved network effects using a linear Instrumental Variable (IV) estimator. This method successfully recovers key interaction features even in finite networks, providing robust tools for analyzing spillovers.
Key concepts
- Generalized Social Interaction ($J$)
- This concept describes how one agent's choice is influenced by others in the network. Unlike simple models, this allows for complex effects where some agents conform to others (positive interaction) while others actively differentiate or oppose them (negative interaction).
- Internal Instrumental Variables
- Because unobserved network factors make direct estimation difficult, the researchers create 'internal instruments.' They randomly split the networks and use the average behavior from one sub-group as a proxy to estimate the effects in another, effectively bypassing measurement error bias.
- Identification Strategy
- This is the mathematical trick used to isolate specific unknown parameters from observed data. By exploiting the panel structure of discrete choice models, they developed a strategy that allows them to recover crucial interaction features ($J$) without needing strong assumptions about unobserved contextual network effects.
Terminology used across episodes
This episode discusses
The paper
Discrete Choice with Generalized Social Interactions · Read on arXiv
Oscar Volpe
Harvard University
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We're looking at the paper: "Discrete Choice with Generalized Social Interactions".
Warren: Here's the gist of the paper. Text A presents an abstract/summary of a theoretical model concerning social interactions in discrete choice models, focusing on identification strategies and empirical applications (Project STAR and deworming uptake).
Ruth: Text B provides a detailed summary of the performance evaluation of an Instrumental Variable (IV) estimator applied to a specific empirical setting—primary school deworming—including robustness checks and misspecification tests.
Warren: I will now synthesize these two distinct pieces of information into one comprehensive, detailed summary, ensuring it maintains the rigor required for high-stakes research.
Ruth: Comprehensive Research Synthesis: Discrete Choice Models with Generalized Social Interactions and IV Estimation Performance This body of work addresses two interconnected areas: the development and theoretical properties of a novel discrete choice model incorporating generalized social interaction effects, and the rigorous empirical validation of an Instrumental Variable (IV) estimator applied to policy spillovers.
Title and authors: Ruth: We just touched on what the paper is about, but let's talk about who wrote it and what they did with this title, Discrete Choice with Generalized Social Interactions.
Warren: The authors are Oscar Volpe and they’re from Harvard University Department of Economics. It sounds like a solid academic foundation for the work they're doing on social choice models.
Iris: They’ve built an equilibrium framework that extends what Brock and Durlauf did before, but the key difference here is allowing those interaction effects to change in sign and size depending on which agent you are looking at.
Ruth: So instead of assuming everyone conforms positively, they let the model allow for both positive spillovers—where people follow their peers—and negative spillovers, where they actively choose to be different.
Warren: That variability is what makes the estimation tricky because you’re dealing with a whole matrix of these effects that need to be identified from individual choice data alone.
Iris: And that's where they introduce this partial identification strategy, which relies on the panel structure of the data itself to control for those unobserved network context things.
Ruth: It seems like they’re using the way agents interact within the same group as a tool to figure out what's actually going on in those underlying social dynamics.
Warren: That sounds like they’re trying to find a way around the problem where you can't observe everyone’s true expectations, which is a big deal in any network model with incomplete information.
The paper's summary: Ruth: The summary of Discrete Choice with Generalized Social Interactions really boils down to them developing empirical tools to study these models where choices have both positive and negative spillovers.
Warren: They build this equilibrium framework that lets the sign and magnitude of interaction effects change across agents in the network, which is a big step up from older models.
Iris: The paper shows how you can recover key economic quantities from this choice data, but it also constructs an estimator specifically designed to estimate those endogenous interaction effects.
Ruth: That estimator is what they claim can be readily applied to standard datasets, which is the practical part of the research for us here on the show.
Warren: The paper then applies these methods to two large-scale education experiments, drawing inferences about how nonuniform interaction effects actually play out in real-world settings.
Iris: They focus specifically on modeling a network of agents who make binary choices subject to private payoffs and social interaction effects, as outlined in Section two of the work <ref:2610.27324#pg2,modeling a network of agents who make binary choices subject to>.
Ruth: And what they found is that differences in social interaction effects—where people conform more strongly to similar peers—actually shape people’s choices, which has direct implications for how we target policy interventions.
Warren: So it’s not just a math exercise; they are linking these theoretical interactions directly to real-world outcomes in education data.
The paper's improvements: Ruth: Now let's look at the specific improvements the authors suggest for this model and what that means for how we think about choice modeling.
Warren: They propose a way to handle those unobserved network effects by exploiting the panel structure inherent in the model, which lets them control for those unobservables by comparing outcomes of different agents in the same network.
Iris: That’s a clever move because it lets them draw inferences about systematic variation in social interaction effects without having to make strict assumptions about those unobserved contextual factors.
Ruth: It also tackles the issue of measurement error bias that comes up when researchers only observe realized choices in finite networks, which is a major identification hurdle.
Warren: To fix that endogeneity, they construct internal instruments by randomly partitioning the network and using the average action in one partition as an instrument for the other.
Iris: That construction allows them to derive a linear instrumental variable estimator that they prove is consistent as the number of networks grows larger.
Ruth: And this consistency proof is important because it tells us that even with these messy finite network conditions, we can still get a reliable estimate of those key interaction features J.
Conclusion: Warren: So to wrap up the paper Discrete Choice with Generalized Social Interactions, the main implication is that we can now identify endogenous interaction effects even in settings where unobserved network context exists.
Ruth: It means we don’t have to impose overly restrictive assumptions about how people interact within a network just to get meaningful results on policy targeting.
Iris: The paper shows that when you allow for both positive and negative spillovers, the welfare dynamics become complex, and those trade-offs are actually identifiable using these new tools.
Warren: In short, they’ve given us a more robust way to estimate how social pressures influence individual decisions in finite networks.
Ruth: It’s a solid paper that moves beyond just looking at simple conformity and allows for the more nuanced reality of people choosing to be different sometimes.
Iris: I just think the consistency proof for that internal instrument construction is what really solidifies their identification strategy here >
More episodes
- Daily Summary for 2026-11-19
- 2608.21274-Recommendation Quality and the Concentration of Consumption: Experimental Evidence from Netflix
- 2408.02757-A nonparametric test for diurnal variation in spot correlation processes
- 2306.16393-High-Dimensional Canonical Correlation Analysis
- 2605.17299-Geometric Brownian motion with intermittent entries and exits
- 2607.04753-Fooling Yourself: how narratives shape beliefs
- 2408.14671-Inference on Treatment Effects with Repeatedly Mismeasured High-Dimensional Controls
- 2201.06898-Difference-in-Differences Estimators for Treatments Continuously Distributed at Every Period
- 2501.15307-Influence Functions: A Unified Differentiation-Based Framework
- The day in conversation for 2026-11-19