PHI403 Lecture 11 — Is More Data Better?

The data-quantity question, reframed as an ontological one. Two views of causation disagree on the answer:

Humean universalism — C causes E iff all instances of C are followed by E. Causation is general and derived from perfect correlations. Hence: more data is better, representative sampling is essential, external validity is achievable, quantitative methods win.
Dispositionalist singularism — each causal set-up has its own set of dispositions and manifestation partners; causation occurs in the single unique instance. Hence: understanding context, mechanism, and individual tendencies matters more than accumulating sample size.

The problem of induction is illustrated by a bag of 1000 marbles — only after examining every marble do we know its colour distribution with certainty. Any pre-complete conclusion is inductively fallible.

Cartwright & Hardie (Evidence Based Policy: A Practical Guide to Doing it Better) supply the key failure case: a nutrition-education-for-mothers program worked in India but not in Bangladesh — because in Bangladesh the mother-in-law cooks the food. Without local causal/contextual knowledge, even a well-run RCT's results do not transfer. This is the canonical statement of the external-validity problem.

The singularist's answer: the problem of induction is not about lacking a perfect data set, but about the ever-present possibility of interference.

Connections

Back-link: sources/2023-anjum-rocca-phi403-causation-in-science. Concepts: concepts/dispositionalism · concepts/regularity-theory-of-causation · concepts/rct-limitations · concepts/causation.