My dissertation is composed of three essays on the economics of education. The first essay, "Teachers' Comparative Advantage, School Segregation, and Educational Mobility in Chicago Public Schools," co-authored with Lauren Sartain, examines a basic assumption that value-added (VA) models make. Specifically, this essay tests whether teacher effects are student specific. We develop and estimate a flexible multivariate VA model in which teacher effects can vary by student type and drift over time. Defining student type by race and using data on 1.7 million observations, we employ quasi-experimental and holdout strategies that exploit teacher switching in Chicago Public Schools. We find evidence of student-specific teacher effects, thus rejecting the homogeneity assumption in VA models. Multivariate teacher effects naturally create comparative advantage (CA)---that is, teachers are more effective with specific student types than other types. We characterize teachers' CA and relate it with educational mobility and racial match effects. We also discuss the implications of deselecting teachers based on univariate rather than multivariate VA when schools are segregated. In future work, we will simulate a policy that matches teachers and classrooms based on CA and document its efficiency and equity impacts on student outcomes. The second essay, "Evaluator Bias: Unpacking the Relationship between Classroom Characteristics and Classroom Observation Ratings," co-authored with Lauren Sartain and Andrew Zou, examines the association between within-teacher changes in classroom composition and teacher observation ratings. We find that having higher-achieving students, less disruptive students, and fewer disadvantaged students, even when the teacher remains in the same school, is associated with better observations. We, then, simulate the distribution of teacher performance under a hypothetical policy that adjusts observation ratings for classroom characteristics, more akin to value-added measures. 30 percent (8.6 percent) of teachers who would otherwise be in the bottom 5th (10th) percentile of the unadjusted rating distribution are above that percentile on the adjusted distribution. The last essay, "The MPACT Initiative: Using Behavioral Tools to Increase Children's Early Math Skills," co-authored with Susan E. Mayer and Ariel Kalil, tests the effectiveness of a low-cost treatment on parental engagement in math activities and a child's early math skills. This chapter specifically asks (i) whether providing parents with information and materials in the form of a math activity booklet increases children's math skills, and (ii) whether a behaviorally informed treatment designed to overcome present bias and increase parents' use of the materials affects children's math skills beyond providing the information and materials alone. MPACT is a 12-week randomized control trial with more than 1,400 parents of preschool-age children attending 29 Head Start Centers in the City of Chicago. We collected data from parent surveys, teacher surveys, time-preference tasks, and children's math test scores at baseline and immediately following the 12-week intervention. Child assessments in addition are collected at 6 and 12 months postintervention. The results in this chapter are for short-term treatment impacts (12-week). Results show an increase in children's engagement with math activities at home but no positive treatment impacts on test scores at 12 weeks. Heterogeneity analysis suggests that Spanish-speaking families benefit more from the intervention than other parents. We discuss possible explanations for the absence of positive treatment effects, including spillover effects.