Partial identification allows applied researchers to learn about parameters of interest without requiring them to make assumptions that guarantee point identification. This course will offer applied researchers an introduction to partial identification and its use in empirical work in economics. No prior knowledge of partial identification is required. Students should however be familiar with commonly used econometric methods such as ordinary least squares, two stage least squares, and maximum likelihood.
As an introduction, the course will begin with a review of point identification and the derivation of estimating equations in familiar contexts, such as the classical linear model. We will then illustrate how the same deductive logic can sometimes result in partial identification. A key area of focus will be on models that produce moment inequalities.
We will then review several areas of economics in which partially identifying models have been applied, such as the study of treatment effects, models with missing data or censored variables, and instrumental variable models with discrete outcomes. We will discuss the features of data and the models used across different applications to produce empirical results.
Available methods for performing estimation and inference will be demonstrated, using a combination of STATA, R, and MATLAB. Some familiarity with each of these will be helpful, but advanced expertise is not required.