2-Day Course: Causal Inference with Graphical Models
When? September 19-20, 2014, 9 am to 4 pm (both days)
Where? UCLA, Carnesale Commons, 251 Charles E Young Dr West, Los Angeles, CA 90095
This course is part of a week-long program preceding the 2nd Annual BayesiaLab User Conference at UCLA (separate registration required).
This course offers an applied introduction to directed acyclic graphs (DAGs) for causal inference from observational data. DAGs are powerful yet intuitive tools for solving complicated causal problems. The two primary uses of DAGs are (1) determining the identifiability of causal effects, and (2) deriving the testable implications of a causal model. DAGs are also useful for illuminating the causal assumptions behind widely used statistical estimation techniques. This course introduces the essential elements for causal reasoning with DAGs and exemplifies these insights with social and biomedical examples.
Topics include: non-parametric identification by adjustment; d-separation; the difference between overcontrol bias, confounding bias, and selection bias; covariate selection in observational research; causal assumptions in regression; instrumental variables; and recent developments in causal mediation.
Throughout, this course focuses on spotting causal opportunities and understanding causal problems. However, it is not a course on statistical methods.
As this course is part of a practitioner-oriented conference program, special emphasis is given to practical exercises. Thus, students get hands-on experience with numerous examples that implement DAGs in the form of Bayesian networks with the BayesiaLab software platform. In this context, course participants have access to a 30-day license of BayesiaLab Professional Edition.
Felix Elwert, Ph.D. (Harvard 2007), is the Vilas Associate Professor of Sociology at the University of Wisconsin–Madison (curriculum vitae & selected publications). He is an expert in causal inference and regularly teaches courses on the subject. He conducts substantive research on topics in social demography, social stratification, and human mortality. His work has appeared in the American Journal of Sociology, the American Sociological Review, the American Journal of Public Health, and Demography.
This course is aimed at applied researchers with an interest in causal inference from observational data. Thus, the course is relevant for many fields of study, including social and behavioral sciences, analytics, data mining, economics, econometrics, epidemiology, medical research, market research, marketing science, statistics, etc.
Participants should have a good working knowledge of applied regression analysis. Some prior exposure to the counterfactual approach to causal inference and basic probability theory will be helpful but are not essential. Participants need to bring their laptop computers to participate in the software-based exercises. Prior familiarity with BayesiaLab is a plus, but is not required.
Location & Venue
UCLA, Carnesale Commons, 251 Charles E Young Dr West, Los Angeles, CA 90095