“Minimally biased” g-loadings of crystallized and non-crystallized abilities

Gignac [Gignac, G. E. (2006). Evaluating subtest ‘g’ saturation levels via the single trait-correlated uniqueness (STCU) SEM approach: Evidence in favor of crystallized subtests as the best indicators of ‘g’. Intelligence, 34, 29–46.] used a single-trait correlated uniqueness (STCU) CFA approach to calculate “minimally biased” g-loadings of the subtests of Wechsler-derived intelligence batteries. On the basis of results showing that g-loadings were higher for crystallized than for non-crystallized abilities, Gignac concluded “in favor of crystallized subtests as the best indicators of ‘g’”. In this article, we show that this conclusion is incorrect. First, we demonstrate using simulated data that STCU models having very good fit to the data of small variable sets can nevertheless calculate subtest g-loadings that are substantially biased. Second, we show that analyses of large and diverse variable sets do not reveal crystallized subtests to have the highest g-loadings.