For nearly two decades, nudges—subtle changes to choice architecture that steer people toward "better" decisions without restricting options—have been among the most popular policy tools in behavioral economics. Default enrollment in retirement plans, simplified nutritional labels, and warning messages on unhealthy products have been adopted by governments worldwide. The assumption has been straightforward: if nudges move people toward choices that experts consider better, they must improve welfare. Allcott, Cohen, Morrison, and Taubinsky (2025), publishing in the American Economic Review, challenge this assumption with a framework and experimental evidence that suggest the answer is more complicated than the nudge literature has acknowledged.
The Core Argument
The paper, titled "When Do Nudges Increase Welfare?", makes a conceptual point that should have been obvious but was largely absent from the literature: a nudge that moves the average choice in the "right" direction can still decrease welfare if it increases the variance of choice distortions.
What does this mean in practice? Consider a fuel economy label on a car. If some consumers are already well-informed about fuel costs and the label nudges them to overweight fuel economy relative to other attributes they care about, the label has increased their choice distortion even while correcting the distortion of other, previously uninformed consumers. The net welfare effect depends on the balance between these two groups—and this balance is an empirical question, not a theoretical certainty.
The Framework
The authors adapt public finance sufficient statistic approaches to characterize nudge welfare effects. The framework yields several key insights:
Average effects are not enough. Most nudge evaluations report the average treatment effect—the nudge reduced sugary drink purchases by X%, or increased fuel-efficient car purchases by Y%. But welfare depends on both the mean and the variance of effects across consumers. A nudge that reduces average distortion but increases variance can be welfare-reducing.
Prices and taxes matter. When markets have pre-existing taxes (fuel taxes, soda taxes) or endogenous prices (car prices that respond to demand), the welfare effect of a nudge interacts with these instruments. With optimal taxes already in place, the nudge's average effect becomes irrelevant—only the variance effect matters. With zero pass-through of taxes to prices, the nudge's effect on consumer choice has no welfare consequence because prices do not adjust.
Heterogeneity is the key. The welfare effect of a nudge depends on who is affected and how. If the nudge primarily corrects the choices of consumers who were most distorted, it improves welfare. If it primarily distorts the choices of consumers who were already choosing well, it reduces welfare.
The Experimental Evidence
The authors implement their framework with two field experiments. The first evaluates automotive fuel economy labels. Enhanced labels shift purchases toward more fuel-efficient vehicles on average, but the distribution of effects reveals that some well-informed consumers overweight fuel economy relative to other attributes they care about. The net welfare effect is ambiguous.
The second experiment evaluates health warning labels on sugary drinks. Labels reduce purchases on average, but some consumers who were already moderating their intake reduce further (past their own preferred level), while the heaviest consumers are least affected. Accounting for this heterogeneity, the labels may decrease aggregate welfare.
Critical Analysis
<| Claim | Source | Assessment |
|---|---|---|
| Nudges can decrease welfare by increasing choice distortion variance | Allcott et al., 2025 | Supported — formal framework with experimental validation |
| Average treatment effects are insufficient for welfare evaluation | Allcott et al., 2025 | Supported — this is a mathematical result, not an empirical claim |
| Fuel economy labels may have ambiguous welfare effects | Allcott et al., 2025 | Supported — experimental evidence, though sample-specific |
| Sugary drink labels may decrease aggregate welfare | Allcott et al., 2025 | Plausible — depends on welfare function specification |
| Optimal taxes make nudge average effects irrelevant | Allcott et al., 2025 | Supported — follows from the sufficient statistics framework |
Implications for Behavioral Public Policy
The paper does not argue that nudges are bad. It argues that nudges are not automatically good—and that the standard evaluation methodology (measuring average behavior change) is insufficient for determining whether they improve welfare.
This has several implications:
Evaluation standards must change. Showing that a nudge "works" (changes average behavior) is not enough. Policymakers need to estimate the distribution of effects across consumers to assess welfare.
Targeting matters. Nudges that reach primarily the consumers who need correction are welfare-improving. Nudges that are applied uniformly—like mandatory labels on all products—inevitably affect some consumers who do not need correction, potentially reducing their welfare.
Interaction with other policies: In markets already subject to corrective taxes (fuel taxes, sugar taxes), the marginal welfare contribution of an additional nudge is smaller and may be negative. Policy instruments should be considered as a portfolio, not individually.
The Broader Behavioral Economics Debate
Lades, Zawojska, Johnston, and colleagues (2025), in the Review of Environmental Economics and Policy, argue that many "anomalies" in stated preference data are not errors but expected behaviors reflecting genuine features of human decision-making. This complements Allcott et al.: if preferences are inherently heterogeneous and context-dependent, nudges that assume a single "correct" choice direction are conceptually problematic.
Colin-Jaeger and Dold (2025) push the critique into political philosophy, arguing that nudge-based policies raise autonomy questions that welfare calculations alone cannot resolve.
Open Questions
Where This Stands
Allcott et al.'s paper is a corrective to the nudge literature's tendency to equate behavior change with welfare improvement. The finding that well-intentioned nudges can reduce welfare by increasing choice distortion variance is both technically sound and practically important. It does not invalidate the nudge approach, but it raises the evidentiary bar: demonstrating that a nudge changes behavior is no longer sufficient. Future nudge evaluations must also demonstrate that the behavior change improves welfare across the distribution of affected consumers.
Explore related work through ORAA ResearchBrain.