Journal article: When choices are inconsistent due to behavioral biases, there is a theoretical debate about whether it is necessary to impose the structure of a model in order to provide precise welfare guidance based on those choices. To address this question empirically, we use standard data sets from the lab and field to evaluate the predictive power of two conservative “model-free” approaches to behavioral welfare analysis. We find that for most individuals, these approaches have high predictive power, which means there is little ambiguity about what should be selected from each choice set. We show that the predictive power of these approaches correlates highly with two properties of revealed preferences: the number of direct revealed preference cycles and the fraction of revealed preference cycles that are direct.