Accelerating the development of formal thinking in middle and high school students IV: Three years after a two-year intervention

Same cohort as Adey & Shayer 1990 — ~24 classes in 9 schools, two schools excluded from analysis; GCSE results for the 11+ starters

England; students who began CASE aged 11 in Year 7, followed to GCSE at age 16.

This is the paper that made the far-transfer claim famous, and everything in the archive's scepticism toolkit fires at once. Non-randomised matched classes; developer-led; two non-implementing schools already dropped; a small number of classes; and the headline effects are SUBGROUP effects that flip sex between cohorts — in the earlier 12+ cohort the boys showed the GCSE gains in science and maths, in this 11+ cohort "the effect was limited to girls." The outcome measures are, unusually and to the authors' credit, genuinely independent: nationally set and marked GCSE examinations, three years after the intervention ended. But the authors themselves report that some of the achievement gain in science, maths and English was NOT associated with gains on the Piagetian tests — i.e. the claimed mediator does not mediate — and attribute it post hoc to "metacognition". Note the sequence: no science-achievement effect at end of treatment (Adey & Shayer 1990), a science effect one year later (Shayer & Adey 1992), and far transfer to maths and English three years later. A treatment effect that grows for five years while its proposed mechanism fails to mediate is the shape of a selection artefact, not of a durable cognitive change.

The single strongest far-transfer claim in science education. Three years after a two-year CASE intervention delivered inside science lessons, GCSE effect sizes versus control classes were science 0.67σ, mathematics 0.72σ and English 0.69σ — i.e. transfer to two other school subjects at roughly the same magnitude as the home subject, on national examinations the developers did not write. The effect was limited to girls in this 11+ cohort, having been limited to boys in the earlier 12+ cohort. Half the students showed science gains consistent with the Piagetian model; for the rest, achievement gains in science, maths and English were NOT accompanied by Piagetian gains above controls.

High. Non-randomised matched classes in volunteer schools, with non-implementing schools removed from analysis; three-year-delayed outcomes give ample room for differential attrition, setting, option choice and school effects. Any intake or school-quality difference between arms is fully expressed in GCSE results.

Failed. The independent randomised effectiveness trial of the same approach at scale (EEF/Hanley et al. 2016, 53 schools, 8,000+ pupils) found g = -0.01 on science and NEGATIVE point estimates on standardised English (-0.15) and maths (-0.11) — opposite sign, same construct, vastly better design. The developers' remaining defence is horizon: the EEF trial measured at end of Year 8, whereas this claim is about GCSE three years later, and the promised GCSE follow-up of the EEF cohort has not been published. Shayer (1999) and Shayer & Adhami (2007) report further positive results, but from regression-to-national- norms designs with self-selected adopter schools, not controlled comparisons.