The Analytics Power Hour #285: Our Prior Is That Many Analysts Are Confounded by Bayesian Statistics
28 snips
Nov 25, 2025 In this discussion, Michael Kaminsky, Co-CEO of Recast and a specialist in Bayesian statistics, offers intriguing insights into the complexities of Bayesian analysis. He emphasizes how human beings naturally think in Bayesian terms, often updating beliefs with new evidence. Kaminsky contrasts Bayesian and frequentist statistics, pointing out the limits of traditional methods. He advocates for involving domain experts to set realistic priors and explores the application of Bayesian modeling in real-world scenarios, including pandemic and decision-making contexts.
AI Snips
Chapters
Books
Transcript
Episode notes
Model The World, Then Compare
- Bayesian statistics model the world as simulations and compare those simulations to observed data.
- This lets analysts infer parameters by matching model implications to real observations.
Bayes Works Without Random Samples
- Bayesian methods naturally handle problems that don't fit the random-sample-from-population framing.
- They let you ask forward-looking questions like forecasting revenue without forcing a sampling metaphor.
Four-For-Four Baseball Example
- Michael Kaminsky used a baseball example: a batter goes 4-for-4 on opening day and you must predict season average.
- Bayesians combine that small sample with prior knowledge of typical batting averages to form a reasonable estimate.
