Applied Linear Modeling

Catalog description

This course deals with how to critically and responsibly model real-world data to answer social science, education, and social policy-related questions, using the framework of the general linear model. Linear modeling (which, in statistics, is synonymous with regression analysis) is the workhorse of much of quantitative social science and, despite its enormous flaws and powerful limitations (which this course will also cover!), it remains an important tool to understand and be able to use. The course builds up multiple regression from correlation and bi-variate regression, and then covers categorical independent variables, nonlinear transformations and polynomial terms, diagnostic checks, model-building and model iteration, interaction effects, mediation analysis, and logistic regression. Mathematical (e.g., Gauss-Markov) assumptions are covered but the emphasis is on deeper epistemic assumptions and more immediate practical limitations. While not covered in detail, pointers will be given to techniques for specific types of data (especially multilevel modeling for nested data) and to important modern developments (especially structural causal modeling, non-parametrics, and machine learning). Throughout, the course will return to and emphasize critiques of linear modeling, to encourage students to be able to use (or choose not to use and oppose) regression analysis rigorously, critically, and responsibly. The course will be taught using R. This course includes an introduction to R. Background in R or in programming is not strictly necessary for this course but it is helpful. Prerequisite: MSSP 6300 Quantitative Reasoning/Social Statistics, or another Introductory graduate statistics course.