This paper argues that standardized tests measure not only what students know, but also whether they know what they know and what they do not know---i.e.,their metacognition. I develop a framework that recovers metacognition from test-taking choices. When answering is costly---e.g., under penalties for errors---students with good metacognition should omit the questions they are least likely to answer correctly, improving their accuracy. Applying the framework to a large field experiment, I find that students largely fail this benchmark, because many students sort questions nearly at random and a sizable minority sort them adversely. Yet, metacognition varies substantially in the population, with systematic gender differences. Using a structural model, I quantify these patterns and decompose scores into knowledge and metacognition components. The findings suggest that interpreting scores as signals of knowledge alone entails sizable information loss, and that metacognitive differences help explain why gender score gaps widen with ability.

Paper download

For further information please contact: erika.somma@unibocconi.it