Scientific knowledge suppresses but does not supplant earlier intuitions

150 college undergraduates verifying 200 statements across 10 knowledge domains

US college undergraduates recruited from introductory psychology, having taken an average of 3.1 college-level maths and science courses (SD 3.3). Adults only - hence off-scope for an archive covering ages 4-18, but recorded in full because it is the single most load-bearing result on whether taught science actually replaces the naive conception.

Within-subjects speeded-reasoning task, read in full text. Participants verified 200 statements as fast as possible - 20 in each of 10 domains (astronomy, evolution, fractions, genetics, germs, matter, mechanics, physiology, thermodynamics, waves), 5 concepts per domain. Half the statements were CONSISTENT (same truth value under both the naive and the scientific theory, e.g. "The moon revolves around the Earth"); half were INCONSISTENT (truth value reverses across the conceptual change, e.g. "The Earth revolves around the sun"). Materials were counterbalanced three ways: equal numbers of objectively true and false statements per domain; relational complexity matched across statement type; word count held constant across statement type, response type and domain. This is the design that makes the alternative explanations testable, and the paper tests and rejects them: syntactic difficulty (matched by construction), and familiarity (the effect ran OPPOSITE to the familiarity prediction - participants were slowest on inconsistent items in the MOST familiar domains). The logic is direct: if instruction overwrites the naive theory, inconsistent statements should be no harder than consistent ones, because the naive theory is no longer there to conflict.

Adults with years of science education were significantly LESS ACCURATE on inconsistent than consistent statements, F(1,149) = 1582.49, p < .001, with the difference robust in every one of the 10 domains (all within-domain t(149) > 5.0, p < .001). They were also significantly SLOWER on inconsistent statements, F(1,149) = 349.98, p < .001, again in every domain (all t(149) > 3.0, p < .01), and the response-latency gap held for 43 of the 50 individual concepts. Mean latency difference was 415 ms on false items and 428 ms on true items - so it is not a response bias. The gap was 562 ms in domains where overall accuracy was 80% or greater, 388 ms at 70-80%, and 334 ms at 70% or below; the linear contrast of speed on accuracy was F(1,149) = 23.40, p < .05. THE DECISIVE DETAIL FOR THIS ARCHIVE: participants with GREATER domain expertise showed MORE cognitive conflict, not less - individual accuracy correlated positively with the consistent/inconsistent latency gap in 6 of 10 domains (fractions r = .27, genetics r = .19, germs r = .27, matter r = .21, physiology r = .22, thermodynamics r = .18, all p < .05), and participants in the top accuracy quartile were 465 ms slower on inconsistent items than those in the bottom quartile. Learning the science does not remove the intuition; it adds a competitor that must be overridden, and the better you know the science the more visibly you are overriding. The effect appeared in domains where the scientific concept is taught in early childhood (fractions, germs, physiology) as well as in late adolescence (evolution, mechanics), so it is not an artefact of recency. The authors state the direct challenge: these findings "present a strong challenge to standard models of conceptual change", including coalescence-and-differentiation and ontological-reassignment models, because the obsolete intuitions keep exerting influence.

Not applicable - within-subject comparison of two statement types in the same people.

Consistent with and extends earlier single-distinction findings (teleological reasoning under speeded conditions in adults, living/non-living categorisation, re-emergence of teleological thought in Alzheimer's patients). Independently corroborated by neuroimaging in a different lab and paradigm: Masson et al. 2014 (experts recruit inhibitory regions on misconception-relevant electricity items) and Potvin et al. 2020 (chemistry professors, longer RTs plus inhibitory-control activation on incongruent items).