Description: Linear predictor as approximation to conditional expectation function. Least-squares projection as sample counterpart. Splines. Omitted variable bias and panel data. Bayesian inference for parameters defined by moment conditions. Finite sample frequentist inference for the normal linear model. Statistical decision theory and dominating least squares with many predictor variables; applications to estimating fixed effects (teacher effects, place effects) using panel data. Asymptotic inference in the generalized method of moments framework. Likelihood inference using information measures to define best approximations within parametric models. Instrumental variable models and the role of random assignment; applications include models of demand and supply and the evaluation of treatment effects.
Course Notes: Enrollment is limited to PhD students in the Economics Department, Business Economics program, and PEG program. Other students wanting to enroll in the course should contact the instructor.
Recommended Prep: probability at the level of Statistics 110; linear algebra.