In this section, we introduce an approach to making such inferences via potential outcomes. We'll discuss approaches to a number of data collection designs and associated problems commonly encountered in clinical research, epidemiology, economics, etc. Topics considered include the fundamental role of the assignment mechanism, in particular, the importance of randomization as an un-confounded method of assignment, and methods for limiting bias in the analysis of observational data, including propensity score analysis.
What am I going to get from this course?
- Understand sources of bias in nonrandomized (and even randomized) studies, and how to use directed acyclic graphs (DAGs) and propensity scores to address.
- Similarly, understand how sampling and selection bias can arise, and how to use Heckman (two-stage) modeling to address.
- Finally, some interventions don't have a control (like a hospital-wide intervention) and are assessed by comparison to historical controls. One way to assess is through interrupted time series, or discontinuity regression models.
- For all the above, participants will be able to apply (SAS) software solutions, de-bug/work out problems, and interpret in a causal context.