John Worrall (2007), Philosophy Compass, 2(6), 981–1022 — DOI-10.1111/j.1747-9991.2007.00106.x
Key takeaways
- The EBM evidence hierarchy’s privileging of RCTs lacks a sound general rationale. Of five standard arguments for randomisation, four fail (Fisher’s significance-testing logic, control of unknown confounders, observational exaggeration, probabilistic causality). The fifth, selection bias, is “cast-iron”, but it supports only taking allocation out of the experimenters’ hands, which randomisation is one way to do.
Seeded from the abstract and a skim, before the reading in NOTE-126. What follows is what the work says about itself.
Medicine, like any rational activity, obviously has to rest on evidence; the interesting question is how general principles of evidence apply there. Worrall focuses on the Evidence-Based Medicine movement and raises difficulties for the rationales given for its evidence hierarchy, especially the privileged place of RCTs and meta-analyses of RCTs. He does not oppose a scientific approach to medical evidence. His aim is to show where philosophers of science can help implement that approach properly.
Standing in the record
Filed from the 2026-09-26 sweep of the papers-feed tracker for philosophy: 20 seconds of active reading over 1 session. It was filed Deferred, unread. NOTE-126 is the close reading of 2026-09-26, and it placed the work: Active — worth reading as the fullest single statement of the philosophical case against the special epistemic standing of randomisation. It is clearly argued and well documented, and it explicitly concedes the selection-bias argument. It is a partisan position piece, and its defenders-of-randomisation opponents (Pocock, Elbourne, Papineau) are summarised, not engaged at length.
When seeded, before the reading in NOTE-126, the entry gave a priority for a deeper reading: medium — 20 s dwell, long and foundational, with §4 carrying detailed arguments the skim does not reproduce.
What the seeding asked a deeper reading to check (NOTE-126 is that reading):
- A classic challenge to the “gold standard” status of RCTs, directly relevant to how the anthology weighs randomized ablations and A/B tests against observational comparisons.
- Check §4(b) in full: the unknown-confounders argument is the one most often used in ML experiment design (random seeds, splits).
- Read with Cartwright (LIT-213) and Reiss (LIT-118), which extend the anti-hierarchy line. Check later responses defending randomization (e.g. on selection bias and allocation concealment).
Access when seeded: compass.onlinelibrary.wiley.com returned 403. I read the published version (Compass typesetting, 42 pp) from a course-reading PDF at joelvelasco.net, found via web search and fetched with curl; the host is a third-party teaching page, not the publisher. Crossref gives online 2007-11-14. Read the abstract, §1 opening, all section and subsection heads, the openings of §3 and §4, and the Conclusion. Keywords are mine.