In this course, students learn how to answer basic causal questions using the multivariate linear model: posing the question, writing the model, selecting the data, estimating, and testing. Accompanying the theory, the course also trains students on searching for data sets, reading codebooks, investigating how the data was created, as well as data cleaning and estimation using Stata. There is a strong focus on analyzing the limitations of the assumptions involved: what could be wrong, what can and cannot be done to fix problems, and how to report results taking in consideration the flaws of the study.
This is a second year graduate course designed for applied students. The course has five parts.
- Nonparametric estimation using kernel methods, including simple kernel regression and local polynomial regression.
- Series methods, using basic expansions, splines, and wavelets.
- Semi-parametric models, including partially linear, separable, and index models.
- Topics: endogeneity in nonparametric models, regression discontinuity design, and use of a nonparametric plugin for the propensity score.