Beyond g: Putting multiple intelligences theory to the test

200 adults (116 women), two tests per hypothesised intelligence domain plus the Wonderlic Personnel Test as the g marker

Mostly undergraduates at Brock University (183) and Western Ontario (17); age 17-66, mean 22.7. Out of the 4-18 age band; psychometric rather than instructional test

The only direct psychometric test of MI's core structural claim, which is INDEPENDENCE of the eight intelligences. Two tests were SELECTED for each of Gardner's domains to match his own descriptions - mostly pre-existing standardised instruments (ETS Kit of Factor-Referenced Cognitive Tests, O'Sullivan & Guilford social-intelligence tests, GATB Mark Making, Gordon's AMMA, the Stork Stand balance test), not instruments built for the study - then factor-analysed by PCA, principal-axis factoring and maximum likelihood, which agreed. Cross-sectional and adult, hence grade D and an off-scope population - retained because the instructional claim rests on the structural claim, and no comparable child study exists. Recorded as type 'quasi-experiment' only because the controlled vocabulary has no cross-sectional/psychometric category; there is no treatment. Full-text verification was done against Visser's MA thesis of the same title, which is the source document for the published Intelligence article and carries the same N and the same loadings table.

Only one large factor emerged (first eigenvalue 3.74, 23.4% of variance; a clear break only between factors 1 and 2). g-loadings (principal-axis) ranged from .03 to .75 and tracked how cognitive the test was, not which "intelligence" it belonged to: Naturalistic .75/.57, Logical-Mathematical .70/.24, Spatial .55/.50, Linguistic .54/.50, Interpersonal .53/.36, Intrapersonal .27/.16, Musical .18/.10, Bodily-Kinaesthetic .06/.03. Over 70% of the 45 inter-test correlations among cognitive abilities were significant and positive - a positive manifold, which is what MI theory denies. Residual within-domain correlations after removing g ranged from -.04 (Intrapersonal) to .29 (Linguistic), so the domains resemble the group factors of standard hierarchical models rather than independent intelligences. The 'intelligences' that escape g are the ones involving sensory, motor or personality content. Visser et al. concluded the findings "contradict Gardner's assertion that there are at least eight independent intelligence domains".

Not applicable in the causal sense; this is a measurement-structure finding. It bears directly on the archive's premise, since a strong g factor across diverse abilities is what the hereditarian literature also finds.

The non-independence finding has been reproduced with different instruments and with children: Almeida et al. 2010 ran a CFA on Gardner-style MI tasks plus a general aptitude battery in 294 children and found a single MI factor (i.e. the intelligences are not separable), and Castejon et al. 2010 found no model fit well but the best-fitting structure was one built on intercorrelation among the intelligences. Note the qualification: Almeida found no common factor spanning the MI tasks and the aptitude battery, so the g-dominance half of Visser's result is less cleanly replicated than the non-independence half. Plucker et al. 1996 is the contrary case, reporting separable linguistic, logical-mathematical and spatial subscales. Gardner (2006) rebutted Visser by arguing the measures leaned on linguistic and logical-mathematical ability; Visser et al. (2006b) replied that intercorrelation depends on individual differences in performance rather than item format, and asked Gardner to supply standard measures. None exist to this day.