Statistical Modeling for Policy Analysis
Statistics is easy. Good statistics – the kind you’d be comfortable basing public policy on – takes a little more work. This class focuses on the intuition underlying statistical technique, the art of learning from others’ mistakes, and the practical challenges of estimating, trouble-shooting, comparing, and interpreting models using real data. Students will estimate statistical models and report on their results throughout the course; they will also compile data, estimate and fine-tune a model, and draw policy implications for a final paper and presentation on a policy issue of their choice. Specific topics will include the following: The role of uncertainty and modeling in policy choice; The classic ordinary least squares (OLS) model; Specification issues, including omitted variable bias, heteroskedasticity and autocorrelation, variable selection, and functional form; Causality, endogeneity, and instrumental variables; Research design, including experiments and quasi-experiments; and Combining and incorporating results from previous studies. We will also consider some of these (somewhat more advanced and specialized) subjects: Limited dependent variables, including Poisson, binary and multiple response, Tobit, and survival analysis; Time-series (ARIMA) and forecasting models; Multivariate time-series, the error-correction model, and cointegration; Panel data and hierarchical linear models; Matching, regression discontinuity, and interrupted time-series designs; and Robust alternatives to OLS. Textbooks Jeffrey Wooldridge, Introductory Econometrics: A Modern Approach (Cengage, 2014). Peter Kennedy, A Guide to Econometrics (MIT Press, 2008) (recommended). Software Stata, available from www.stata.com. Buy Stata/IC 14. A perpetual license will cost $198; six-month and annual licenses are also available.