Implicit Theories of Intelligence Predict Achievement Across an Adolescent Transition: A Longitudinal Study and an Intervention

Study 1 n=373 US seventh graders followed to eighth grade; Study 2 n=91 (48 intervention, 43 control) seventh graders at the same school

One New York City middle school, relatively low-achieving; US seventh graders.

The paper the whole applied literature descends from, and it is much smaller than its influence suggests. Study 1 is purely correlational growth-curve modelling. Study 2 randomised 91 individual students within maths classes to eight 25-minute weekly sessions delivered by trained mentors, with an ACTIVE control of equal dose (study skills plus a lesson on memory instead of the malleable-brain lesson). No preregistration, no blinding stated, no confidence intervals reported anywhere, and the outcome is teacher-assigned maths grades - the standardized citywide achievement test appears only as a covariate, never as an outcome.

Study 1 - incremental theorists showed an upward maths-grade trajectory and entity theorists a flat one (theory x time t = 2.93, p < .05); the path coefficient of theory on grade change was beta = .20 (p < .05) and dropped to .11 (not significant) once motivational mediators were added. Study 2 - the manipulation worked (between-group post d = .47, F = 4.50, p < .05) and condition predicted change in maths grades across the intervention (b = .53, t = 2.93, p < .05), with no effect on grades before the intervention (b = .18, ns) or on the preceding interval (b = -.12, ns). The condition-by-baseline-mindset interaction, on which the targeting logic rests, was only marginal (b = -.28, t = -1.71, p < .10).

Low for the randomised Study 2 contrast; Study 1 is observational and its association is fully open to the archive's default explanation.

Ganimian 2020 used a single-session adaptation of this exact intervention in 202 Argentine schools and found a precisely estimated zero. The English EEF effectiveness trial found zero on national tests. Under the archive's replication rule an n=91 result with grades-only outcomes that fails at scale cannot support more than a mixed verdict.