A common approach to estimating whether a general (G) factor helps explain the relations among various specific (S) factors, is bi-factor modeling. In a symmetrical bi-factor model, each item loads onto its given S factor and the broader G factor. However, symmetrical bi-factor models are inappropriate unless the subscales can be construed as randomly drawn from a broader array of interchangeable S factors (Eid, Geiser, Koch, & Heene, 2017). Bi-factor-S-1 models circumvent this issue by leaving one S factor un-modeled, which allows that un-modeled S factor to conceptually define the G factor (Heinrich, Zagorscak, Eid, & Knaevelsrud, 2018).
