Many important questions about cause and effect are impractical to answer with a randomized experiment. What should we do instead? In this episode we talk about doing causal inference with observational data. Has psychology's historical obsession with internal validity led it, ironically, to think about causal inference in an unsophisticated way? Can formal analytic tools like directed acyclic graphs (DAGs) tell us how to do better studies? Or is their main lesson don't bother trying? How do norms and incentives in publishing help or hurt in doing better causal inference? Plus: We answer a letter about applying to psychology grad school when your background is in data science.
- Thinking Clearly About Correlations and Causation: Graphical Causal Models for Observational Data, by Julia M. Rohrer
- That one weird third variable problem nobody ever mentions: Conditioning on a collider, by Julia Rohrer
- The selection-distortion effect: How selection changes correlations in surprising ways, by Sanjay Srivastava
The Black Goat is hosted by Sanjay Srivastava, Alexa Tullett, and Simine Vazire. Find us on the web at www.theblackgoatpodcast.com, on Twitter at @blackgoatpod, on Facebook at facebook.com/blackgoatpod/, and on instagram at @blackgoatpod. You can email us at firstname.lastname@example.org. You can subscribe to us on iTunes or Stitcher.
Our theme music is Peak Beak by Doctor Turtle, available on freemusicarchive.org under a Creative Commons noncommercial attribution license. Our logo was created by Jude Weaver.
This is episode 76. It was recorded on March 16, 2020.