Demonstrating the feasibility of automating experimental bias evaluations within probabilistic programming frameworks
Date:
Javier Mencia Ledo, Feifan Liu, Scott Schwartz.
Undergraduate research presentation.
This talk built on Dr. Michael Evansβ framework to quantify statistical evidence using relative belief ratios (RB), addressing estimation (π) and hypothesis testing (π). Key axioms (minimality, bias control, falsifiability) guide ingredient selection (e.g., priors, models). Bias types (against and in favor) are defined for π/π and computed via probabilistic programming (PyMC) for toy/regression models, leveraging marginal likelihoods and Cholesky decompositions for scalability.
Click Relative_Belief (1).pdf to view it