The United States educational policy requires that K-12 students participate in annual standardized tests. As a result, school districts that have traditionally utilized ongoing “formative" assessments of student progress, are increasingly relying on additional, costly “interim" assessments. In addition, some districts are experimenting with merit-based incentives that tie teachers' bonuses to student performance on state tests. We examine the relationship between information on student performance and monetary incentives for teachers using a two-period principal-agent model.

A wide range of operations problems are built on an underlying probabilistic model. However, estimation error always exists when learning from these models, which can lead to misleading decision-making. Moreover, these models are inherently mis-specified to a certain degree, which calls for robust learning and policies for these operations problems. In this talk, we will discuss robust machine learning and its applications to operations problems.