Optimal College Dropout with Correlated Ability Signals

Abstract

Students learn their true ability during college by receiving grade signals. However, students are uncertain of how informative or noisy these signals are. Sequentially re-evaluating the decision to drop out or remain enrolled in college, a student’s incorrect understanding of signal structure can bias ex-ante expected returns to degree completion, resulting in nonoptimal dropout timing. In this paper, I explore correlated signal noise as a factor that may influence dropout behavior. First, I present empirical evidence that suggests students’ dropout decisions respond to signal noise. These findings motivate a model in which signal noise is correlated, influencing the optimal time of dropout. Structural estimation of this model weakly suggests that students behave as if signal noise correlation is zero. To better understand such naivety, I propose an estimation procedure allowing for unobserved heterogeneity in correlation perception.