PHI403 Lecture 12 — The Explanatory Power of Mechanisms

The course's treatment of mechanisms and their relationship to statistical evidence.

Statistical vocabulary routinely smuggles causal assumptions into correlational conclusions — "increase", "prevent", "affect", "reduce", "enhance". Without a mechanism, we cannot license interventions based on a statistical finding. "Married men live longer" is statistically true but the real causal work is done by stress, diet, life-style, loneliness, depression — a marriage can affect all of these in either direction.

The Russo-Williamson thesis: causal claims require both statistical and mechanistic evidence. In conflict, favour statistics. The lecture cites the Semmelweis case as a cautionary tale: his hand-washing hypothesis was rejected because no mechanism of disease transmission was known, even though the statistics were striking.

Causation of stability: equilibrium (Moon's orbit) and homeostasis (body temperature, blood-oxygen, insulin regulation) are instances of causation where nothing changes — counterbalancing powers keep the system in a range. Change-based and difference-making theories struggle here; a dynamic theory of mechanism is needed.

The lecture reprints the EBM evidence hierarchy — RCT > observation > mechanism — and criticises the ranking on mechanism's behalf.

Connections

Back-link: sources/2023-anjum-rocca-phi403-causation-in-science. Concepts: concepts/mechanisms-in-science · concepts/evidence-hierarchy · concepts/causation · concepts/dispositionalism.