Data Science, Statistics, Research Methods, Psychological Science, Epidemiology
do-Calculus for Humans Pearl’s structural approach to causal inference is built around DAGs (directed acyclic graphs). In many cases, applying this framework boils down to measuring and adjusting for confounding variables. A key insight is that when working with purely observational data, controlling for all confounders can be equivalent to assigning the treatment level, as we would in an…
Causal Inference Is Hard The first post argued that causal inference is simply a matter of ruling out rival explanations. That sounds pretty straightforward, but it’s not. Rival explanations can be hard to identify, hard to measure, and hard to eliminate. Overlook a confound, measure it poorly, or choose an inadequate study design, and causal inference fails. No matter how sophisticated the…
Causal Inference Is Easy Causal inference seems to be having a moment. Why the sudden interest? The tech industry’s belated realization that correlation really doesn’t equal causation, no matter how big the data, could have something to do with it. It might also be that we’ve collectively worked through all the easy prediction problems and now need to answer harder questions about why things…
CUPED: Old Wine in a New Bottle? Running AB tests in industry comes with several challenges. Detecting small but meaningful effects is difficult and time consuming. You can run the experiment longer, recruit more users, or accept that some effects will go undetected. None of those options are appealing when you’re running lots of tests and your stakeholders wanted the results yesterday. Deng et…
Abelson’s Laws Robert Abelson was a faculty member in the Yale Psychology Department for almost 5 decades. He did trailblazing research that helped lay the foundation for the cognitive revolution in psychology and was early to adopt computer programming to model social cognitive processes. He also worked with John Tukey on his “Swing-o-metric” technique for predicting election results, becoming…