Measurement Asymmetry in g-Achievement Research: A Construct Validity Analysis of GPA as Criterion

Grade Point Average (GPA) is the dominant criterion variable in research on the relationshipbetween general intelligence (g) and scholastic achievement. The most rigorous meta-analysis todate (Roth et al., 2015) estimated the population correlation at ρ = .54 after correcting for samplingerror, predictor unreliability, and predictor range restriction. This paper argues that an analogouscorrection programme has not been applied to the criterion side, and that GPA fails theconstruct-validity standards routinely applied to intelligence tests. Specifically: GPA's latentvariable is unspecified and locally variable; its inter-rater reliability is substantially below that ofthe predictor; its variance is substantially driven by executive functioning, behavioural compliance,and conscientiousness rather than academic ability alone; and its measurement properties arenon-invariant across grading contexts. Under the causal validity account of Borsboom,Mellenbergh, and van Heerden (2004), GPA is not a valid measure of academic ability. Theresulting measurement asymmetry — a construct-validated predictor correlated against aconstruct-unvalidated criterion — means that ρ = .54 estimates the relationship betweenintelligence and a contaminated composite, not between intelligence and academic ability. Thisasymmetry is diagnostically visible in neurodivergent populations (ADHD, ASD, twice-exceptionallearners), where systematic IQ–GPA discrepancies mediated by executive functioning andbehavioural compliance are more parsimoniously interpreted as criterion-validity failures than asevidence of underachievement. Implications for g-achievement research methodology, educationalassessment, and educational policy are discussed.