Cattell–Horn–Carroll models of intelligence frequently show that, at the group-factor level, G f is most strongly related to g , whereas at the subtest level, G c -associated measures exhibit the highest g loadings. One proposed solution to this “ g paradox” holds that Stratum-III g and Stratum-II G f are identical, and that the sizeable g loadings of crystallized subtests merely reflect the investment of G f into learning. Investment theory is weakly evidenced, however. We argue that the “ g paradox” results from subtests measuring facets of G f exhibiting pronounced specificity for cognitive entities. Capturing everything that goes into G f is difficult on a single-measure basis, hence lower Stratum-I g loadings. The G f group factor is nonetheless reflective of the composite of these entities and therefore is uniquely (at Stratum II) associated with g . Subtests measuring G c broadly index the quality of global systems involving many cognitive processes, not entities, and so relate to factors that have formative effects on g , which are Stratum-I specific. We posit the existence of two distinguishable sources of general covariance: a formative g (associated primarily with G c ) and a reflective g (associated primarily with G f ), with the latter hierarchically superordinate to the former. Network analysis of “pure” psychometric measures of the G c and G f domains indicates that the former exhibits significantly greater network integrity than the latter, consistent with this formative/reflective model. Random effects meta-analysis of SEM contrast parameters, derived from four large genetically informed studies, finds that subtests assigned to a “ G c ” category are associated with higher-magnitude direct (formative) genetic paths relative to those in a “ G f ” category, suggesting a weak but discriminable and broad Stratum-I g in the residual covariance structure. Given the theorized phylogenetic histories of these two g s, we term the formative (“bottom-up”) g “proto g ” ( g p ), and the reflective (“top-down”) g “neo g ” ( g n ).
Sarraf et al. (Wed,) studied this question.