Description
The gold standard of scientific inquiry is a Randomized Controlled Trial (RCT). Random assignment cures a variety of ills in experimental design.
Unfortunately in the social sciences, an RCT can often be cost-prohibitive, impossible to construct, or unethical. In disciplines like Epidemiology and Economics, the solution has been observational studies. These studies provide some understanding of the issues explored, but lack the causal assertions of an RCT. Luckily, we don’t have to give up there. This talk focuses on techniques that we can use to draw rigorous conclusions from data we already have. We will explore propensity matching and related tools to estimate the probability of treatment for a given subject. We’ll look at matching, and explore implementation details for making this algorithm (much) more efficient. And we’ll finish with a discussion of the limitations/sensitivity of the results. Attendees will walk away with a tool set for drawing more robust conclusions from data in the messy world of observational studies.