Reading Statistical Rethinking Through the Statistical Bridge

Abstract

Richard McElreath's Statistical Rethinking teaches Bayesian modeling through a wider discipline of scientific reasoning. The golem metaphor limits the authority of statistical machinery. Small worlds and large worlds separate formal coherence from scientific adequacy. Generative reasoning asks how observations could arise. Causal analysis begins from scientific structure. Measurement and missingness keep the route from world to data visible. Version 1.0 of this paper drew one general proposition from those features: assumption and inference are passage-relative roles. Version 2.0 keeps that proposition and reads it through The Statistical Bridge Version 3. The mature Bridge distinguishes bridge constitution, probability source, inference certificate, evidential status, and claim license. These distinctions make several parts of McElreath's pedagogy easier to inspect. A likelihood can supply an evidence-side probability source. A prior can supply target-side probability. A posterior can function as an inference certificate. A causal premise can support identification while retaining its own evidential status. The paper now has a narrower job. It concentrates on explicit models, generative reasoning, small-world discipline, causal structure, measurement, and executability. Broad school comparisons belong to Where Does Probability Live?. Iteration, checking, and the loop/tangle distinction belong to the later reading of Bayesian Workflow. The resulting interpretation is compact. A model can serve as a premise in one passage and become an object of criticism in another. A causal graph can support an estimand and later become the target of a sensitivity analysis. A dataset can serve as evidence and later become the object of a measurement investigation. Epistemic role can change while evidential standing stays fixed. Warrant changes when relevant evidence changes.