Learning Bayesian Statistics #6 A principled Bayesian workflow, with Michael Betancourt
12 snips
Jan 3, 2020 The podcast discusses a principled Bayesian workflow with Michael Betancourt, highlighting the challenges of building models and the importance of questioning default settings. Michael shares insights on Bayesian vs. frequentist methods in inference, mastering the Bayesian workflow, diverse projects in the Stan team, and personal endeavors. The episode also covers custom model building, upcoming courses on advanced topics, and resources for Bayesian methods.
AI Snips
Chapters
Transcript
Episode notes
Probabilistic Modeling Confusion
- People often confuse excitement for Bayesian methods with excitement for probabilistic modeling.
- Frequentist methods can be model-based but are usually presented as orthodox estimators.
Frequentist Uncertainty
- Frequentist methods, when used correctly, also quantify uncertainty, contrary to popular belief.
- Maximum likelihood analyses often confuse people due to their focus on point estimators, obscuring the presence of uncertainty.
Bayesian Simplicity
- Bayesian methodology simplifies implementation because it frames everything as probabilistic calculations.
- Frequentist methods, requiring calibration against counterfactual data, are arguably more complex.