SDS 607: Inferring Causality — with Jennifer Hill

Basic Concepts

Jennifer Hill introduces the fundamental concepts of Bayesian statistics, contrasting them with frequentist approaches. She emphasizes the flexibility and honesty of Bayesian methods in expressing uncertainty and avoiding overfitting. Jon Krohn adds that Bayesian statistics allow for more adaptable modeling strategies, which can be particularly useful in complex data scenarios 1. Hill also discusses the importance of prior specification in Bayesian models to prevent overfitting and improve model performance 2.

In Bayesian statistics, we think parameters are unknown and uncertain, and we put a prior on them to express our uncertainty about them.

--- Jennifer Hill

Hill's pragmatic approach to Bayesian statistics highlights its practical advantages in real-world applications.

   

BART Tool

Hill delves into the advantages of Bayesian Additive Regression Trees (BART) for causal inference. She explains that BART, unlike other model fitters, provides coherent uncertainty estimates and avoids overfitting through clever prior specification 3. This makes BART particularly effective in scenarios where randomized control trials are not feasible. Jon Krohn notes that BART's flexibility and Bayesian framework make it a powerful tool for causal inference 4.

The nice thing for me is that it's embedded in a Bayesian framework, so you get a bunch of bonuses for that.

--- Jennifer Hill

Hill's insights into BART underscore its utility in making reliable causal inferences in complex data environments.