February 2021
We introduce a method for experimentally evaluating interventions designed to improve the quality of choices in settings where people imperfectly comprehend consequences. Among other virtues, our method yields an intuitive sufficient statistic for welfare that admits formal interpretations even when consumers suffer from biases outside the scope of analysis. We use it to study a financial education intervention, which we find improves the quality of decisions only when it incorporates practice and feedback, contrary to the implications of analyses based on conventional efficacy metrics. We trace the failures of conventional metrics to violations of assumptions that our method avoids.