This dissertation is composed of three distinct chapters, each of which addresses issues of estimating treatment effects. The first chapter empirically tests the Value-Added (VA) model using school lotteries. The second chapter, co-authored with Michael Wood, considers properties of inverse probability weighting (IPW) in simple treatment effect models with selection bias. The third chapter, co-authored with Guanglei Hong and Heather D. Hill, explores unbiased estimation of treatment effect models with mediators. VA is designed to estimate the treatment effects of schools and teachers on students' academic outcomes. They rely on a conditional independence assumption that is not testable within-sample: students are not sorted to schools based on unobserved components of their test score. There has been an extensive debate in the economic and education policy literature regarding this issue but very few empirical tests, especially for school--as opposed to teacher--VA. This chapter directly tests the assumption using school lotteries, in which students are randomly assigned an offer from a charter school. First, I formally link VA to a model for the Intent to Treat (ITT) effect of winning the lottery. Next, I show how the ITT parameter can be expressed as a linear combination of parameters from VA. I then replace those parameters with estimates from estimating the VA model on the full sample of students and schools in the district, and show that this VA-based estimate of the ITT parameter will be unbiased if the VA school effect estimates are themselves unbiased. My results fail to reject the null hypothesis that estimates from VA are unbiased. I test a number of different VA models used frequently in the literature, and also discuss the implications of my results for teacher VA models. The second chapter explores the performance of IPW as a treatment effect estimator in linear models with selection bias. IPW has been presented as an estimator that has the advantage of being non-parametric. However, its advantages over ordinary least squares (OLS) may turn into a disadvantage in the presence of selection bias, which will frequently occur in observational settings. IPW may inflate selection bias by giving the greatest weight to observations that contribute most to bias. We present our intuition, and then provide a proof that verifies this and other conjectures. Using simulations and real data, we compare the bias of IPW, OLS and other re-weighting estimators that we introduce. We allow for heterogeneity in the treatment effect, and suggest how researchers may be able to differentiate between heterogeneity and misspecification. The third chapter introduces and applies an innovative weighting approach to causal mediation analysis. This method allows for a treatment-by-mediator interaction, unlike other techniques for estimating the effects of mediators. Applied to the setting of a welfare-to-work experiment, our weighting method decomposes the treatment effect into direct and indirect effects, and demonstrates how to estimate these effects using ratio of mediator probability weighting. We also employ simulations to test the robustness of our estimators to misspecification of the propensity score model. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]