When the Norm is Wrong: Social Norm Misspecification and Targeting with an Application to Microfinance Adoption
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: We're looking at the paper: "When the Norm is Wrong".
Warren: Here's the gist of the paper. Structural models of peer effects capture peer influence through average, aggregate, or extreme peer actions.
Ruth: This paper extends the constant elasticity of substitution (CES) norm from continuous to discrete choices within a binary-choice game featuring incomplete information.
Warren: It derives results concerning the existence and uniqueness of a rational expectations equilibrium and establishes conditions for identifying structural parameters.
Ruth: The model is estimated using the Nested Pseudo-Likelihood (NPL) estimator, implemented via the spectral algorithm, which addresses convergence issues inherent in the original NPL algorithm.
Paper summary: Ruth: We just talked about how this paper tackles structural models of peer effects by moving from continuous choices to discrete ones using a constant elasticity of substitution norm, and they found that imposing a linear norm biases their estimates even in large samples.
Warren: Right, so the core claim is that the way we capture peer influence matters profoundly, and when you use an incorrect model for that—like assuming linearity instead of something else—your results about who to target get skewed significantly.
Iris: The paper sets out to study this misspecification by extending the constant elasticity of substitution norm into a binary choice game with incomplete information, and they derive conditions for existence and uniqueness of a rational expectations equilibrium.
Ruth: That means they are proving that there is a specific set of conditions under which the model actually works logically, which helps us know what we need to observe in our real data to confirm those assumptions hold.
Warren: And their estimation method uses the Nested Pseudo-Likelihood estimator implemented with a spectral algorithm, which they developed specifically because it handles the convergence issues that pop up with the original pseudo-likelihood method.
Iris: The paper also establishes conditions for identifying those structural parameters, which is crucial because without identification you can't actually use the model to make predictions or policy recommendations.
Ruth: So, the main point they are making is that it’s not enough to just have a peer effect model; you have to specify *how* that influence aggregates in discrete settings for your analysis to be meaningful.
Warren: And they show this isn't just theoretical noise; they tested it on microfinance adoption data in India and found a strongly max-leaning norm with significant heterogeneity in the curvature of those norms.
Iris: That heterogeneity is important because it shows that even within different groups, like caste or wealth segments, the way social norms play out changes substantially.
Ruth: And the empirical results show that this is not just an academic finding; imposing linearity distorts peer mechanisms by more than threefold in terms of strategic complementarity and understating conformity.
Warren: So if you’re running a study on adoption, you need to be careful about your norm specification because a small change there can lead to completely different conclusions about who is most responsive to intervention.
Iris: And the optimal targeting policy derived from their work shows that the CES planner focuses subsidies on village leaders in a max-leaning regime, but this targeting differs significantly if you use a min-like norm instead.
Ruth: It really hammers home that misspecifying the norm or ignoring that underlying heterogeneity can completely misdirect your interventions by targeting the wrong households and forecasting adoption changes wrongly.
Warren: This is about making sure the model reflects reality, not just fitting the data in a way that makes it look nice on a plot.
Iris: And they do flag one limitation: their analysis is specifically for discrete choice settings with both complementarity and conformity, so we can't assume this exact framework applies to all types of social interaction models.
Conclusion: Ruth: So, to conclude on "When the Norm is Wrong: Social Norm Misspecification and Targeting with an Application to Microfinance Adoption," the authors are essentially arguing that the shape of a social norm isn't just a minor detail in modeling peer effects.
Warren: They’re saying that whether you model peer influence as averaging or taking extremes, you have to choose your specification carefully because those choices lead to different policy recommendations.
Iris: The key implication is practical: if you are designing a policy intervention, your choice of norm—whether it's linear or something more complex like the max-leaning one they found—is what determines which households get the subsidy and how effective that subsidy actually is.
Ruth: It shows that ignoring underlying heterogeneity across different segments, like caste or wealth groups, can lead to fundamentally different and often wrong targeting strategies.
Warren: So the authors are telling us that getting the social norm shape right is critical for effective policy recommendations, because a wrong assumption about the norm translates directly into targeting the wrong people.
Shantanu Chadha
University of Warwick
econ.EM
Submitted: 2026-11-09
Updated: 2026-11-09
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 92/100
The gist: Structural models of peer effects capture peer influence through average, aggregate, or extreme peer actions.
Key concepts
- Constant Elasticity of Substitution (CES)
- This model uses a specific mathematical structure to capture how agents evaluate social norms based on peer actions. It assumes a consistent way agents weigh the influence of different peers, allowing researchers to test if this specific norm shape is more realistic than simpler linear models.
- Rational Expectations Equilibrium
- This is the theoretical state where all agents make optimal decisions based on their beliefs about what others will do. The paper derives conditions for when such an equilibrium exists and how to find it, which is crucial for understanding how adoption patterns stabilize.
- Norm Misspecification
- This occurs when a researcher uses an incorrect mathematical model (like assuming a linear norm instead of the CES structure) to describe social influence. The paper shows that this error distorts estimates and leads to flawed policy recommendations, as the wrong assumptions about peer behavior are embedded in the results.
Terminology
Summary
Structural models of peer effects capture peer influence through average, aggregate, or extreme peer actions. p. 0 This paper extends the constant elasticity of substitution (CES) norm from continuous to discrete choices within a binary-choice game featuring incomplete information. p. 0 It derives results concerning the existence and uniqueness of a rational expectations equilibrium and establishes conditions for identifying structural parameters. p. 0 The model is estimated using the Nested Pseudo-Likelihood (NPL) estimator, implemented via the spectral algorithm, which addresses convergence issues inherent in the original NPL algorithm. p. 0
The theoretical foundation involves agents evaluating social norms based on their beliefs about peers' actions, where these beliefs directly enter utility. Under expected utility maximization, agent i compares the expected utilities of two actions. p. 5 The difference in utilities between choosing action 1 and action 0 is shown to be linear in y tilde-i(beta), as the squared norm terms cancel out. p. 1 This leads to an adoption condition where agent i adopts if and only if alpha i + theta 1 d ie E[y tilde-i(beta)] + theta 2 E[y tilde-i(beta)] - 1/2 at least eta i, where y tilde e-i(beta) is the belief about the norm. The paper adopts this belief-based norm because it is tractable, keeps beta informative with binary actions, and possesses a behavioral foundation requiring correct beliefs for equilibrium. p. 8
Empirical application to microfinance adoption data in India reveals that the CES estimates of peer effects converge toward true values as sample size increases, whereas imposing a linear norm biases these estimates even in large samples. p. 0 Specifically, the empirical results reject the linear norm in favor of a strongly max-leaning one. p. 4 Imposing linearity distorts the peer mechanism by overstating strategic complementarity more than threefold and understating conformity. p. 4 The optimal targeting policy derived from the estimated norm concentrates subsidies on village “leaders”—households that are central to information diffusion—whereas imposing linearity targets different households, missing this crucial effect. p. 6
Furthermore, the norm regime exhibits heterogeneity across caste and wealth segments. p. 0 For illustrative segments, the CES planner concentrates the subsidy budget on high-degree households in a max-leaning regime, while under a min-like norm, targeting differs significantly. p. 5 The optimal subsidy allocation under CES and linear norms shows that both policies fund almost disjoint sets of households with negatively correlated rankings; the CES planner favors village leaders (62% of recipients) compared to 12% under the linear norm. p. 5 This demonstrates that misspecifying the norm or ignoring underlying heterogeneity can misdirect interventions, targeting the wrong households and incorrectly forecasting adoption changes. p. 6 The paper concludes that correctly specifying the shape of the social norm is critical for effective policy recommendations. p. 1
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved policy recommendations by correctly identifying household types targeted for subsidies; specifically, systems can now distinguish between
village leaders
favored by a max-leaning norm versus broader sets favored by a linear norm. -
Enhanced estimation accuracy through the use of the Nested Pseudo-Likelihood estimator implemented with the spectral algorithm, which overcomes convergence issues in non-linear systems and provides consistent estimates of structural parameters even when peer effects are complex.
-
Robust identification of social norm parameters; AI systems can now precisely estimate whether a social norm is strongly max-leaning (e.g., estimated beta around 491) or minimum-like (e.g., estimated beta around-283) based on observed adoption patterns in specific demographic segments like caste and wealth.
-
Better handling of network heterogeneity; the system can segment villages along caste and wealth axes to estimate segment-specific norms and peer effects, moving beyond the limitation of imposing a single norm across an entire dataset.
-
More reliable counterfactual analysis; by correctly specifying the discrete choice model primitives, AI systems can predict adoption outcomes under different subsidy budgets with high fidelity, avoiding biases introduced by misspecifying the linear norm assumption.
Sources
- Identification of Heterogeneous Peer Effects
- Quantile Peer Effect Models
- Asymmetries in Peer Effects
- Heterogeneity in peer effects for binary outcomes
- Who Matters to Whom? Identifying Peer Effects with Propagation Geometry
Related papers
- Stationarity with Occasionally Binding Constraints
- Backtesting High-Dimensional Expected Shortfall
- Higher-Order Neyman Orthogonality in Moment-Condition Models
- Online Bayesian Inference for a Dirichlet Process Mixture of Binary Logit Models
- Debiased Conditional Independence Tests for Causal Effects
- Sharp Testable Implications of Encouragement Designs