Workshop:
Introduction to Causal Inference for Time-Varying Treatments in Population Science
Felix Elwert, Harvard University
December 2, 2006, 9:00 a.m.-12:00 a.m.
Topic: This workshop provides a broad conceptual introduction to recent developments in causal inference for time-varying treatments. Much of this material was first developed in biostatistics and epidemiology and holds considerable promise for applications in demography.
The potential outcomes (counterfactual) framework of causal inference is firmly established in the social sciences. Most attention, however, has been focused on estimating causal effects from point treatments (e.g. using propensity scores). While point treatments are important, they are often of limited interest in population science. Time-varying treatments, on the other hand, are frequently of central interest. They arise whenever there are multiple interventions over time, the timing of treatment is determined endogenously, and/or inference concerns causal mechanisms via direct or indirect effects. The potential outcomes framework extends naturally to time-varying treatments and clarifies some long-standing statistical issues along the way.
Topics include:
- Point treatments versus time-varying treatments
- Time-varying confounding
- Why standard methods may fail even in the absence of confounding
- The g-null paradox
- Inverse probability of treatment weighting
- Marginal structural models
The workshop emphasizes a thorough conceptual understanding of the basic challenges and solutions. The exposition will draw on graphical tools of identification (DAGs) and eschew abstract mathematical proofs. Applied examples are drawn from demography and public health.
Prerequisites: The workshop is geared toward applied researchers and advanced graduate students in the social sciences. Prerequisites include a good applied understanding of regression analysis and a basic familiarity with the potential outcomes framework of causal inference at the level of Holland (1986) and Winship and Morgan (1999). No advanced math required.
Holland, Paul W. 1986. “Statistics and Causal Inference.” Journal of the American Statistical Association 81(396): 945-970.
Winship, Christopher, and Stephen L. Morgan. 1999. “The Estimation of Causal Effects from Observational Data.” Annual Review of Sociology 25: 659-706.
Instructor: Felix Elwert holds a Ph.D. in sociology and an A.M. in statistics from Harvard University. He works on the application of causal inference in sociology and demography, interspousal health effects, and the causes and consequences of death and divorce for American families. Elwert has taught workshops on causal inference in various settings, most recently at the ASA 2005 Annual Meeting and the 2006 Penn State Summer Institute on Longitudinal Methods. He is currently a post-doctoral Fellow at Harvard Medical School and will start an appointment as assistant professor of sociology at UW Madison in 2007.