We develop a new methodology to estimate subjective beliefs from hypothetical-choice data. Our identification approach is based on the novel insight that by varying the amount of information on future realizations of stochastic variables, discrete-choice experiments can identify not only preferences but also subjective beliefs. We formally prove this result in a general setting and apply it to design a strategic survey instrument to measure Rajasthani parents’ subjective beliefs over the joint distribution of girls’ age of marriage, education, and marriage match quality. Our approach allows us to quantify the importance of perceived marriage market returns to education and youth, and perform various counterfactual simulation exercises. We find that eliminating the perceived marriage market return to education causes a 60% drop in the number of girls still in school at age 16, and almost none continue their education by age 18. Responses to our strategic survey instrument allow us accurately to predict realized schooling trajectories in follow-up data we collect from the same sample five years after our experimental data collection.

This paper analyzes gender differences in working patterns and wages on Amazon Mechanical Turk, a popular online labor platform. Using information on 2 million tasks, we find no gender differences in task selection nor experience. Nonetheless, women earn 20% less per hour on average. Gender differences in working patterns are a significant driver of this wage gap. Women are more likely to interrupt their working time on the platform with consequences for their task completion speed. A follow-up survey shows that the gender differences in working patterns and hourly wages are concentrated among workers with children.